financial literacy, retirement provisions, and household portfolio
TRANSCRIPT
Financial Literacy, Retirement Provisions, and Household
Portfolio Behavior
Four Empirical Contributions
Maarten van Rooij
ISBN: 978-90-393-4974-8 Cover design: Ridderprint Offsetdrukkerij BV, Ridderkerk Cover photos: Rob Meulemans and Martien de Bock © Maarten van Rooij, 2008
Financial Literacy, Retirement Provisions, and Household Portfolio Behavior
Four empirical contributions
Financieel Alfabetisme, Pensioenvoorzieningen en Beleggingsgedrag van Gezinnen
Vier empirische bijdragen
(met een samenvatting in het Nederlands)
Proefschrift ter verkrijging van de graad van doctor aan de Universiteit Utrecht
op gezag van de rector magnificus, prof. dr. J.C. Stoof,
ingevolge het besluit van het college voor promoties in het openbaar te verdedigen
op vrijdag 9 januari 2009 des middags te 4.15 uur
door
Marinus Cornelia Jozef van Rooij
geboren op 10 juli 1970 te Acht (Eindhoven)
Promotoren: Prof. dr. R.J.M. Alessie Prof. dr. L.H. Hoogduin
To my family
i
ACKNOWLEDGEMENTS
Finishing a large project like writing a PhD-thesis, it is good practice for the author to
express gratitude to all those who contributed directly or indirectly. Naturally, this is a
hopeless exercise as so many persons are eligible for a well-deserved thank you. Hence, I
want to let each of you know in this impersonal manner that I am deeply grateful for all of the
interest and support I have received. Nevertheless, without downplaying the contribution of
others, I take the opportunity to mention a few of you explicitly. The first who vigorously
encouraged me to write a PhD-thesis was Sylvester Eijffinger. Although not successful at the
time, I have to acknowledge that his enthusiasm fuelled the idea to write a PhD-thesis; a
thought that never disappeared completely since then. I am certainly also grateful to De
Nederlandsche Bank as I was very much privileged to have the opportunity to write my thesis
in a stimulating, challenging research and policy environment.
A special thank goes to my two promotors Rob Alessie and Lex Hoogduin who not
only provided me with many helpful comments and suggestions, but also sharpened my views
during stimulating discussions on the research findings and the policy conclusions. In
addition, I have learned a lot on how to use econometric techniques that had started fading
away. I am also deeply indebted to all my co-authors who made the research experience less
lonely and much more enjoyable. Annamaria, Clemens, Federica, Henriëtte, and Rob thanks
for the fun and the smooth cooperation; I have learned a lot from all of you. In addition, I
want to thank the members of the reading committee (Harry Garretsen, Michael Haliassos,
Arie Kapteyn and Clemens Kool) for reading the manuscript carefully and for their kind and
helpful comments; I am very pleased and honored by so many, highly appreciated experts in
my commission. Last but not least I am grateful to Peter van Els, who was always involved in
the background and stimulated and sincerely supported me to walk this road already long
before the Spring of 2005 when the idea of writing a thesis became more concrete.
I owe a lot to my father, mother, and twin sisters, but also to friends, colleagues and
other relatives as they not only make life pleasant but also - without knowing it - often formed
the very biased sample of one observation that I build many theories on (some of which I
pursued in this thesis). As of the very beginning of the PhD-project up until now, Monique,
Anne and Iris (in order of appearance) have fully supported me while inevitably they suffered
most from me writing this thesis. My girls, I will try to make it up to you and hopefully you
will also benefit from the increase, if any, in my own financial literacy.
Maarten van Rooij, Amsterdam, November 2, 2008
iii
CONTENTS
Introduction 1 Paper 1: Risk-Return Preferences in the Pension Domain: Are People Able to Choose?
(published jointly with Clemens Kool and Henriëtte Prast, Journal of Public Economics, 91, 701-722, Copyright 2006 by Elsevier B.V.) 13
Paper 2: Financial Literacy and Stock Market Participation (written jointly with Annamaria Lusardi and Rob Alessie) 45 Paper 3: Financial Literacy, Retirement Planning, and Household Wealth (written jointly with Annamaria Lusardi and Rob Alessie) 91 Paper 4: Choice or No Choice: What Explains the Attractiveness of Default Options? (written jointly with Federica Teppa) 135 Samenvatting (Summary in Dutch) 179 Curriculum Vitae 189
1
INTRODUCTION
3
1. Background and motivation
This thesis is a collection of four papers on household financial behavior. The
landscape for the management of household wealth has changed dramatically in a relatively
short time period. On the one hand, economic welfare is at unprecedented levels and provides
many households with increased opportunities to accumulate savings, to invest their wealth
portfolio and to trade off labor and leisure intertemporally. On the other hand, people are
increasingly expected to take individual responsibility for their economic well-being. At the
same time, the deregulation of financial markets has increased competition between financial
institutions and boosted financial innovations which among others has lead to a continuous
stream of new financial instruments. These developments contribute to the increased interest
of the economics profession in studies on household finance (see for example the Presidential
Address to the American Finance Association by John Campbell (2006)), but also explain its
growing relevance in policy-debates since the effectiveness of fiscal and monetary policy
crucially depends on consumers’ responses.
Life cycle models of consumption and savings behavior have been and still are the
most important departing point for the description of household financial behavior. The
simplest versions of these models predict that households accumulate wealth during their
working career to finance retirement thereafter (Modigliani and Brumberg, 1954; Friedman,
1957). Life cycle models have been made much more realistic since then by incorporating for
example uncertainty, liquidity constraints, and bequest motives (see Browning and Lusardi
(1996) for an overview). The underlying basic assumptions however have remained
unchanged in most models, i.e. consumers are considered as rational agents who collect and
process all relevant information and maximize their lifetime utility. The validity of these
assumptions has been questioned by for example psychologists who argue that consumers use
heuristics and are prone to behavioral biases. At the same time there is evidence that
households make financial mistakes violating the implicit assumption in life cycle savings
models that households possess the necessary skills to behave optimally. The common theme
of the papers collected in this thesis relates to retirement behavior and the role of financial
skills in individual decision-making. More specifically, the papers address pension
preferences of employees and their willingness to exercise investor autonomy as well as their
ability to do so; the measurement of individuals’ financial sophistication and ability, and its
impact on portfolio choice, wealth accumulation and retirement planning; the role of financial
literacy and other determinants of individual decision-making in choice situations with a
default option.
4
The papers take a positive approach. We try to describe and explain household
behavior based on information obtained from specifically designed internet surveys among
the Dutch household panel of CentERdata.1 Where historically economists have been
skeptical towards the usefulness of household surveys and the information content of
subjective questions, they are now widely used and have proven to elicit helpful information
with predictive validity on household behavior.2 Moreover, they are indispensable for
generating information on heterogeneous household preferences and attitudes which are
crucial for understanding individual decision-making.
2. Outline
The first paper, entitled ‘Risk-return preferences in the pension domain: Are people
able to choose?’, investigates pension preferences and investor autonomy of Dutch
employees. The Netherlands is an interesting case study as its pension system provides hardly
any freedom of choice, while in the last three decades countries all over the world have
shifted risk and responsibility from employers towards workers. The United States and the
United Kingdom have for example witnessed a major shift towards Defined Contribution
(DC) retirement plans at the expense of Defined Benefit (DB) plans. New international
accounting standards, the stock market crisis in 2000-2003 and the structural decline in capital
market interest rates have fuelled the debate on whether the prevailing DB system is still
affordable or whether more investment autonomy and risks should be shifted to employees.
We find that a vast majority of Dutch employees opposes to changes that provide them
with more individual responsibility for their pension provisions. These preferences are guided
by their attitude towards risk and an introspection of their own financial skills. Respondents
are highly risk averse, especially in the pension domain, and strongly prefer guaranteed
pension benefits upon retirement. In addition, the average respondent considers himself
financially unsophisticated and is reluctant to take control of retirement savings even when
1 While the surveys are answered via the internet, the access to internet is not a prerequisite in the recruitment phase. If necessary, panel members are either provided with a pc with internet connection or a set-top-box that enables them to participate through their television set. An advantage of web-based interviews is that participants do not feel rushed to provide answers and that the absence of an interviewer increases anonymity and reduces the likelihood of social desirable responses. Although, there is some concern of shortcutting or satisficing behavior (quick, inaccurate responses by less motivated respondents) in internet surveys because the computer design provides additional opportunities for multitasking and quickly skipping from one topic to the next one, laboratory and field experiments by Chang and Krosnick (2003) document evidence of less satisficing behavior in a self-administered computer-based survey mode compared to the administration of interviews over the telephone. 2 See for example Hurd and McGarry (2002), Manski (2004), Donkers and Van Soest (1999), and Kapteyn and Teppa (2002).
5
offered the possibility to increase expertise. An experiment shows that respondents who
initially choose a relatively safe investment portfolio in a hypothetical DC-system are likely
to switch to the more risky median investment portfolio when confronted with the probability
distribution of future income flows. This suggests that respondents indeed lack the financial
skills needed for exercising investor autonomy over pension wealth. While these results might
partly reflect the lack of exposure to self-directed savings plans in the past, they also raise
questions and concerns about the general level of financial literacy.
The second paper, entitled ‘Financial literacy and stock market participation’, aims at
measuring household financial knowledge and cognitive ability. Not only pension decisions
but also many other financial decisions have become more complex due to financial
innovations and an increasing supply of complex products (e.g. the loan market). At the same
time, individuals are increasingly expected to take responsibility for their economic well-
being. While financial skills are a necessary prerequisite to deal with increasing individual
responsibility, we have little knowledge on whether individuals are capable of navigating this
new financial environment. We have designed an extensive list of questions to assess basic
financial literacy related to numeracy and the working of inflation and interest rates as well as
questions on more advanced topics related to financial market instruments (stocks, bonds and
mutual funds). Thereby, our work improves upon previous studies by considering more
refined indices of financial literacy.
Our data show that the majority of households display basic financial knowledge and
have some grasp of concepts such as interest compounding, inflation, and the time value of
money. However, very few go beyond these basic concepts; many households do not know
the difference between bonds and stocks, the relationship between bond prices and interest
rates, and the basics of risk diversification. We also contribute to the methodology of
measuring financial knowledge as we show that there is a lot of noise in the responses to
financial literacy questions, i.e. the wording of the questions is critically important for
measuring financial knowledge and minor variations in the wording of questions may cause
vast changes in response patterns. The sensitivity to the wording of survey questions provides
additional evidence for limited financial knowledge, but it also emphasizes the importance of
testing and validating questions in e.g. pilot versions before fielding the final survey.
We evaluate the importance of financial literacy by studying whether more financially
knowledgeable individuals are more likely to hold stocks. Thereby, we add to the literature
which tries to understand the puzzle of limited stock market participation (Haliassos and
Bertaut, 1995). Standard expected utility maximization models dictate that it is optimal for
6
virtually everyone to invest part of their wealth in stocks if only a tiny fraction. The
motivation is that households facing no equity risk will be better in terms of expected utility
when they participate for at least a small amount in the stock market provided that the equity
premium is positive. Nevertheless, in practice in most countries there is a large majority who
stays away from the stock market (Guiso, Haliassos and Jappelli, 2002). The basic idea that
emerges from the literature is that information and processing costs, including costs to figure
out how to invest and how to monitor advisors and outcomes, play an important role in
explaining nonparticipation. This explanation however is inadequate for wealthy households.
Other studies point to the role of trust and social interactions. A high level of financial
knowledge will contribute to lower information costs and reduce the relevance of barriers to
participation. Indeed, we document evidence that individuals with low financial literacy are
significantly less likely to invest in stocks. We employ the variation in the extent to which
people have been exposed to economics during their education, measuring financial
knowledge that existed prior to investing in the stock market, in order to address the issue of
reverse causality i.e. the possibility that people increase their financial sophistication as a
result of their activities in the stock market.
The third paper, entitled ‘Financial literacy, retirement planning, and household
wealth’, focuses on the effect of financial literacy on household net worth. Thereby, it
investigates the relevance of financial sophistication for household behavior and financial
outcomes from a broader perspective. There is ample empirical evidence that many
households are prone to make financial mistakes and some evidence, although not undisputed,
that financial education fosters savings (see Lusardi (2004) for an overview). The latter
studies however do not particularly focus on whether financial education impacts savings via
an increase in financial sophistication or via other mechanisms. The reported effects on
savings could - at least partly - also stem from the provision of information, the offering of
commitment devices, peer effects, or be due to self-selection into financial education
seminars.
Our estimation results document a statistically and economically significant influence
of financial sophistication on wealth holdings. This is important for public policy, especially
in view of the widespread fear that many households do not save enough for retirement.
Indeed, we show that financial sophistication fosters planning for retirement. Thereby, people
collect and process information on (future) income and expenses, and do the necessary
calculations. This process provides them with information on the required retirement savings,
whereas the related activities might help households to overcome problems of self-control and
7
increase wealth holdings (Ameriks, Caplin and Leahy, 2003; Lusardi and Mitchell, 2007). At
the same time, the fact that financial knowledge increases the likelihood of entering the stock
market, thereby improving opportunities to diversify and taking advantage of the equity
premium, might contribute to better portfolio management and higher wealth as well. We
highlight both these channels as potential mechanisms explaining the positive effect of
financial sophistication on household net worth. An additional finding is that for those who
are relatively confident about their financial skills the likelihood to enter retirement planning
activities is higher. Apparently the extent to which people feel uncomfortable about their
financial sophistication is another element that deters people from making information
intensive decisions. This suggests that, in addition to financial education, efforts to present
choice problems in a clear and easily understandable way could effectively support
households in making complicated decisions.
The fourth paper, entitled ‘Choice or no choice: What explains the attractiveness of
default options?’, investigates the impact of financial literacy on retirement savings decisions
from a different perspective. It is well documented that default options attract
disproportionally many decision-makers. However, from a theoretical point of view, the
framing of decisions, and in particular the choice of the default option, should not matter if
preferences are clearly defined and the cost of switching between alternatives is negligible.
Conventional wisdom states that more freedom of choice is always better as people may
choose to ignore new options. Applying insights from psychological research to economics, it
has been shown that choice stress and information overload may have an important impact on
financial decisions because these factors make default choices more likely. However, there
are also other reasons why decision-makers find default options more attractive such as
inertia, status quo bias, the interpretation of defaults as implicit advice, and procrastination.
To the best of our knowledge however there has not been a study that compared the
relative importance of these explanations empirically. We take the viewpoint that individuals
who defer decisions because of their complexity in one choice situation are also more likely
to display this type of behavior in other domains. Persons who procrastinate on the
cancellation of their subscriptions might as well procrastinate in taking care of their
retirement provisions. We consider several heterogeneous choice situations with very
different characteristics and investigate whether there is a dominant factor explaining default
decisions. From this we draw lessons for important financial choices in general and retirement
decisions in particular.
8
Our results show that procrastination and financial illiteracy are important
determinants of default choices in many choice situations in the Netherlands. The situations
we consider include important financial decisions such as savings for old age and early
retirement and having a will. Furthermore, we also consider issues like organ donation, voting
participation, canceling subscriptions and no consent decisions to block commercial
marketing efforts by mail or phone. In addition, the role of social interactions and social
norms seems to be important in explaining deviations from the default because of the impact
of third party opinions on choice behavior.
This paper also contains a comparable analysis for the US as we had the opportunity to
insert a shortened version of our survey on default choices into the RAND American Life
Panel. Interestingly, procrastination and financial illiteracy come forward as important
explanations for default choices in the US as well, but we do not find a similar role for social
norms and interactions. Moreover, in the US financial literacy seems more important than
procrastination, whereas in the Netherlands procrastination seems to be somewhat more
dominant. While we can only speculate about the sources of these differences - which are
likely to be related to the distinct institutions, culture and traditions – they have important
implications for public policy. In particular the findings suggest that in the Netherlands new
policy initiatives require a relatively larger role for increasing awareness, whereas in the US
emphasizing the provision of information and a clear and simple presentation of decisions
might be the most effective way to proceed.
3. Discussion
While the four papers address several dimensions of financial behavior and individual
decision-making, the overall picture that emerges is, first, that financial skills are crucially
important for household decisions and, second, that there is a wide gap between the public’s
actual financial knowledge and ability on the one hand and the financial skills needed for e.g.
retirement savings decisions on the other hand.
This said, the empirical results also make clear that households are not completely
irrational. Dutch employees for example seem to be aware of the lack of financial literacy
which explains why they are not keen to exert more control and investor autonomy in the
pension domain. In addition, while default options do affect individual decision-making, most
variation in choices is yet explained by heterogeneity in objective personal characteristics and
circumstances. At the same time, individuals are more likely to hold substantial wealth
holdings, to take active decisions and to behave according to the standard models of optimal
9
economic behavior when they are better educated, and in particular have higher financial
skills.
A direct implication is that increasing the level of financial education stimulates wise
economic behavior and helps consumers taking their responsibility in important financial
decisions. It is clear that in the last say thirty years, the relative importance of financial skills
has increased and it is important that school curricula reflect this development. An important
topic for further research is how financial education can be organized most effectively also as
regards the timing of activities (both during school and thereafter; e.g. training on the job), the
type of information as well as the way it is conveyed possibly dependent on the target group
(oral or written information, using internet, television or other communication channels, etc.).
Another important question not addressed in this thesis is the relative importance of schooling
and learning by doing in accumulating knowledge and acquiring financial skills. Intuitively,
both seem important and are likely to reinforce each other. Nevertheless, the size of the gap
between financial literacy and the necessary skills to navigate through the current financial
environment suggests that financial education alone will most likely not suffice to close the
gap. Moreover, the results from studies focusing on the US, a country which historically has
been characterized by much freedom of choice and individual responsibility, evoke modesty
as regards to what can be accomplished by learning from experience. Against this
background, the current financial crisis has shown that also highly educated experts with a
huge amount of experience apparently had trouble in understanding complex financial
instruments and the basics of risk management.
The implications for modeling household behavior are less straightforward. The
conclusions involve a clear violation of the basic assumptions underlying standard life cycle
models of utility maximization in which households collect all relevant information and are
equipped with the necessary knowledge and skills to solve complex mathematical problems.
Because of this lack of financial sophistication, it may not come us a surprise that these
models have difficulties in accurately describing actual household behavior. Somewhat more
successful are behavioral models (including models of bounded rationality or satisficing
behavior in which households search for solutions which are adequately appropriate) which
usually focus on specific heuristic behavior, e.g. based upon concepts such as myopia, loss
aversion, mental accounting, or habit formation. While these models might be more
successful in describing household behavior in specific situations, they do not offer a guide on
how households should behave nor offer a comprehensive tool to explain household behavior
in a multitude of situations. At the same time, our research documents widespread differences
10
in financial literacy and heterogeneity in individual decision-making which makes it unlikely
that it is possible to find a synthesis; a comprehensive model incorporating all household
characteristics which can deal with both optimal (prescriptive) and actual (descriptive)
behavior in many situations.
Standard economic models do however provide an important benchmark for optimal
consumer behavior from a rational, economic point of view. The necessity of interventions in
the decision making process of households should be based on how far household decisions
are away from optimal behavior and a careful evaluation of how this situation might be
improved. Policy measures may take the form of information provision and financial
education initiatives to improve financial literacy without impairing the decision autonomy of
households. Interventions may occasionally also take the form of changing default options in
the interest of the average consumer, in regulation and supervision of financial markets, or at
the extreme in limiting the choice options for households. As regards the latter one could
indeed make a case that, given the complexity of pension decisions and the potential
consequences of serious financial mistakes for retirement savings, government intervention in
the pension system is justified (as is done in many countries in the form of the provision of a
state pension or with compulsory participation in company pension plans). Behavioral models
contribute to this process because they give insight in why people show non-optimal behavior
and provide information that policy-makers or regulators can use to create tools or guidelines
for effective interventions. At the same time, one should be very careful with intervening in
individual decision-making as interventions might entail unintended consequences.
Competition in markets where financial products are developed and sold fosters
innovative solutions that are tailor-made for real household financial problems, but at the
same time these markets should be organized in a way to avoid incentives that work against
the interest of financially illiterate consumers. A low level of financial literacy in combination
with typical behavioral traits (procrastination, myopia, inertia, etc.) make consumers
sometimes easily attracted to decisions that are not in their best interest (e.g. going for short
term gains, underestimating or ignoring risks that are involved, etc.). At the same time
financial institutions might be faced with the temptation to exploit these behavioral traits in
view of profit maximization or competitive pressures. This may take the form of the provision
of information which is heavily emphasizing the advantages of products for the customer,
downplaying or ignoring the disadvantages or seducing consumers with special offers to
profit from those who get locked in. Ironically, well-informed consumers take advantage of
the opportunities offered by these selling strategies and might in the end be subsidized by
11
illiterate consumers paying too high prices. Gabaix and Laibson (2006) show theoretically
how this situation can persist while none of the participants (buyers and sellers) has an
incentive to increase transparency and educate the illiterate consumers.
Principally, financial advisors and intermediaries play an important role in dealing
with financial illiteracy. They are able to collect and process information on products and
their conditions from a large amount of financial institutions efficiently and translate them
into easily accessible choices households have to make. The problem however is how to
create incentives to assure that advisors act in best interest of their customers. The same lack
of financial expertise and behavioral traits make consumers vulnerable for financial mistakes
in basing their decisions on financial advice of intermediaries (e.g. the attractiveness of
seemingly free advice, a focus on low interest rates in complex mortgage products, the eye-
catching figures of investment returns based on favorable assumptions). In fact, consumers
need a lot of financial skills to choose independent advisors who act in their clients’ interest
and to be able to understand the consequences of the financial advice given.
Here is a role for regulation to promote transparency and complete and easily
accessible information to ease the consumers’ decision-making process as well as to
safeguard that institutions who operate at the edge of an integer policy towards the
consumers’ interest are not driving out of the market the financial firms that do act in the best
interest of the consumer for a fair price. Helpful measures that are actually gaining popularity
are rules for transparency on all costs involved by financial products and advice in prescribed
formats, transparency on incentive structures, or license systems for organizations and fulfill a
number of condition to secure quality
The whole palette of imposing rules on transparency, information provision and
incentive structures, the use of well-considered designs for decision problems (including the
default option) and in exceptional situations limiting the freedom of choice might moderate
the adverse consequences of limited financial skills. Naturally, the trade-off between
enforcing rules, limiting freedom of choice and using default options to steer consumer
decisions on the one hand and the decision autonomy of households and neutrality of
government institutions on the other hand is a political decision. Nevertheless, given the
empirical evidence on the wide heterogeneity in financial skills and the current lack of basic
economic insights among a substantial group of most illiterate consumers, this type of
measures might be necessary to supplement financial education initiatives as in the end we
cannot expect each consumer to be skilled as an economics expert.
12
References Ameriks, J., A. Caplin, and J. Leahy, 2003, Wealth accumulation and the propensity to plan, Quarterly Journal of Economics, 118, 1007-1047. Browning, M., and A. Lusardi, 1996, Household saving: Micro theories and micro facts, Journal of Economic Literature, 34, 1797-1855. Campbell, J., 2006, Household finance, Journal of Finance, 61, 1553-1604. Chang, L., and J. Krosnick, 2003, Comparing oral interviewing with self-administered computerized questionnaires: an experiment, mimeo, Ohio State University. Donkers, B, and A. van Soest, 1999, Subjective measures of household preferences and financial decisions, Journal of Economic Psychology, 20, 613-642. Friedman, M., 1957, A Theory of the Consumption Function, Princeton University Press. Gabaix, X., and D. Laibson, 2006, Shrouded attributes, consumer myopia, and information suppression in competitive markets, Quarterly Journal of Economics, 121, 505-540. Guiso, L., M. Haliassos, and T. Jappelli, 2002, Household Portfolios, MIT Press, Cambridge. Haliassos, M., and C. Bertaut, 1995, Why do so few hold stocks?, Economic Journal, 105, 1110-1129. Hurd, M., and K. McGarry, 2002, The predictive validity of subjective probabilities of survival, Economic Journal, 112, 966-985. Kapteyn, A. and F. Teppa, 2002, Subjective measures of risk aversion and portfolio choice, mimeo, RAND. Lusardi, A., 2004, Saving and the effectiveness of financial education. In: O. Mitchell and S. Utkus (eds.), Pension Design and Structure: New Lessons from Behavioral Finance, Oxford University Press, 157-184. Lusardi, A., and O. Mitchell, 2007, Baby boomers retirement security: The role of planning, financial literacy and housing wealth, Journal of Monetary Economics, 54, 205-224. Manski, C., 2004, Measuring expectations, Econometrica, 72, 1329-1376. Modigliani, F., and R. Brumberg, 1954, Utility analysis and the consumption function: An interpretation of cross-section data. In: K. Kurihara (ed.), Post-Keynesian Economics, New Brunswick, New Jersey, Rutgers University Press, 388-436.
13
PAPER ONE *
RISK-RETURN PREFERENCES IN THE PENSION DOMAIN: ARE PEOPLE ABLE TO CHOOSE?
(Co-authors: Clemens Kool and Henriëtte Prast)
* Published in April 2007 by Elsevier in the Journal of Public Economics (See Van Rooij, M., C. Kool, and H. Prast, 2007, Risk-return preferences in the pension domain: Are people able to choose?, Journal of Public Economics, 91, 701-722). I am grateful to both Elsevier and the Journal of Public Economics for the kind permission to use this material in my thesis.
15
Risk-Return Preferences in the Pension Domain: Are People Able to Choose?
Maarten van Rooij (De Nederlandsche Bank and Netspar) Clemens Kool (Utrecht School of Economics and Netspar)
Henriëtte Prast (De Nederlandsche Bank, Netspar and Tilburg University)*
August 2006 Published in the Journal of Public Economics in April 2007 **
.
Abstract This paper presents new evidence for the Netherlands on pension preferences and investor autonomy in the pension domain using a representative survey of about 1000 Dutch citizens. Our main conclusions are the following. Risk aversion is domain dependent and highest in the pension domain. The vast majority of respondents favors the currently dominant defined benefit pension system. If offered a combined defined benefit/defined contribution system, the majority would like to have a guaranteed pension income of at least 70% of their net labor income. Self-assessed risk tolerance and financial expertise are important individual explanatory variables of pension attitudes. The average respondent considers himself financially unsophisticated and is reluctant to take control of retirement savings investment, even when offered the possibility to increase expertise. Respondents who have chosen a relatively safe portfolio tend to switch to the riskier median portfolio when they are shown future income streams. This again suggests that many respondents currently lack the skills to have investor autonomy over investment for retirement. Key words: Investor Autonomy, Pension Preferences, Risk Tolerance, Defined Contribution Schemes JEL Classification: D12, D80, G11, J26
∗ Maarten C.J. van Rooij, Economics & Research Division, De Nederlandsche Bank, P.O. Box 98, 1000 AB, Amsterdam ([email protected]), Clemens, J.M. Kool, professor of Finance and Financial Markets at the Utrecht School of Economics of Utrecht University ([email protected]), and Henriëtte M. Prast, professor of Personal Finance at Tilburg University on a Rabo/Netspar chair ([email protected]). The authors are grateful to two anonymous referees, to Rob Alessie, Dimitris Georgarakos, Lex Hoogduin, Kristin Kleinjans, James Poterba (editor), Fred van Raaij, Susann Rohwedder, Margreet Schuit, Paul Sengmüller, Arthur van Soest, Peter Vlaar, Peter van Els, and to participants of the 7th DNB Annual Research Conference on Household Behavior and Financial Decision Making (Amsterdam, November 2004), the 6th RTN Conference on Economics of Ageing (Frankfurt, May 2005) and seminars at DNB, Utrecht University and Netspar, for their valuable comments and suggestions. Moreover, we thank the staff of CentERdata and in particular Corrie Vis for their assistance in setting up the survey and the field work. The views expressed in this paper are those of the authors and do not necessarily reflect those of the Nederlandsche Bank. ** I am grateful to Elsevier, publisher of the Journal of Public Economics, for the kind permission to use this material in my thesis. The reference to the article is: Van Rooij, M., C. Kool, and H. Prast, 2007, Risk-return preferences in the pension domain: Are people able to choose?, Journal of Public Economics, 91, 701-722.
16
1. Introduction
In the industrialized world, risk and responsibility in retirement plans are increasingly
shifted from employers towards workers. Twenty-five years ago the majority of US private
pension plans was purely defined benefit (DB). This number has decreased to 1 out of 10
retirement plans and nowadays the majority is defined contribution (DC) only. The same
trend has appeared in many European countries, with the United Kingdom as the most notable
example. However, some countries, e.g. the Netherlands, are still dominated by DB pension
plans. Nevertheless, even these countries gradually are and will be affected by the trend
towards DC systems.
This paper summarizes and discusses the key findings of a survey on risk-return
preferences of Dutch employees in the pension domain. The focus is on whether people are
able to choose, that is whether individuals have well-defined preferences when it comes to
their pension investments. DB retirement plans are convenient for participants because they
delegate a number of choices on e.g. risk-return considerations of pension investments. A DC
system has the advantage of creating the possibility for individually tailored pension plans.
However, individuals may not benefit from autonomy because of a lack of financial
sophistication (Lusardi and Mitchell, 2005), self control problems, and psychological biases
(Benartzi and Thaler, 2002; Thaler and Benartzi, 2004). Many DC pension funds in the US do
in fact express doubts about the quality of the investment choices made by their participants
(Benartzi and Thaler, 2001; Mitchell and Zeldes, 1996). Choi, Laibson, Madrian and Metrick
(2004) claim that people seem to be aware that their level of retirement saving may be too
low.
Our study contributes to the existing literature on pension preferences and investor
autonomy in a number of dimensions. First, we use a representative sample of the Dutch
population, whereas most empirical studies are based on US higher educated and/or higher
income categories. Second, the respondents in our sample are not used to DC retirement
plans. Third, in the Netherlands, retirement age is not at the discretion of the individual
worker. Fourth, the social security system in the Netherlands is relatively generous, with a
state pension of over € 600 per month for single persons aged 65 and older. Fifth, the
respondents in our panel have recently experienced a serious stock market decline, whereas
those in previous studies were interviewed when the stock market was still booming.
Our main results paint a consistent picture. Dutch employees prefer the status quo of a
DB scheme with a limited say, at most, about the level of pension savings and risk-taking. In
case of a changeover to an individual DC system, only a minority would choose investor
17
autonomy. Given the prevalence of DB in the Netherlands this is likely to be partly due to a
status quo bias, but it is also in line with the fact that employees (correctly) have strong
doubts about their financial skills and report a high level of risk aversion in the pension
domain. If respondents had to choose their retirement portfolio in a DC scheme, they would
be inclined to opt for low-risk, low-return, in line with their risk attitude. When confronted
with two different expected retirement income streams, they tend to choose the riskier of the
two, even when the other portfolio is the one they had initially opted for. This result may be
due to myopic loss aversion and/or to unrealistic asset market expectations. Either way, the
results confirm the doubts expressed by respondents about their ability to take control over
their pension savings. The paper is structured as follows. The next section describes current
pension arrangements in the Netherlands. Section 3 explains our dataset and methodology. In
Section 4, the survey results are presented and discussed. Section 5 provides a summary of the
results and some policy implications.
2. Retirement plans in the Netherlands
The pension system in the Netherlands consists of two pillars.1 The first pillar is a pay-
as-you-go state pension of about € 632 per month for every person aged 65 and over. The
second pillar is capital funded and linked to employment contracts. Employees are obliged to
join their employer’s pension scheme and enroll automatically. Collective insurers and
especially pension funds are responsible for the organization and administration of a large
majority of the pension schemes. The balance total of all pension funds amounts to more than
€ 600 billion, exceeding the gross domestic product of the Netherlands. These assets are
invested in equities (40-45%), fixed income assets (40-45%), real estate (10%) and other
investment categories including commodities.
Until recent times, virtual all Dutch pension schemes in the second pillar had a DB
character and were based on final pay. Recent developments such as the emergence of new
international accounting standards and the ageing of the population have already caused some
changes. Increasingly, defined benefit schemes are based on average career wages. Also,
several corporate pension funds in the Netherlands replaced their DB systems by (collective)
DC arrangements in 2005, while others have expressed their intention to do the same in the
1 In addition, individuals can make private arrangements for further retirement schemes. This facility is of marginal importance in the current system.
18
future.2 About 85% of the 6 million pension fund participants now have DB retirement plans
based on either final pay or career average wages. Roughly 200000 employees (3%) have DC
pension plans. The remaining combined pension schemes typically are DB-type plans with
some DC-elements.
The typical employee in the Netherlands now has a career average defined benefit
pension with indexation conditional on asset-liability ratios. During the active working period
accrued pension rights are in many cases indexed to negotiated wage increases (without
backloading accruals for career steps) and pension benefits are often indexed to consumer
price inflation. However, full indexation of pension claims to cost-of-living increases is not
guaranteed. Consequently, typical Dutch DB retirement plans de facto contain a DC element.
Until recently, though, indexation cuts were very rare. Cost-of-living indexation is (was)
generally financed out of excess returns on pension investments. Since nominal liabilities are
calculated using nominal bond rates, the more risky – equity – composition of investment
portfolios makes average excess returns likely in the longer run.
DB plans in the Netherlands are relatively safe for its participants for a number of
reasons. First and foremost, risk is limited through the system’s funded nature. Moreover,
pension fund supervision is strict and targeted on the prevention of underfunding. Funds are
required to have asset portfolios of at least 105% of their nominal liabilities. In the event of
underfunding, a pension fund has to submit a plan to remedy this within a year. In addition,
the recovery plan should make clear how the fund will reach a capital funding ratio –
including a buffer to deal with disappointing asset market developments – within 15 years.
The equilibrium funding ratio for an individual pension fund depends on market interest rates
and on the composition of the pension fund’s investment portfolio. It currently falls between
120% and 130% for most pension funds. As a result of its emphasis on solvency, the Dutch
pension system has proved to be able to withstand high losses on equity investments. For
example, during the stock market decline in the period 2000-2003 the asset-liability ratio
based on a fixed actuarial interest rate of 4% decreased from 150% in 1999 to 109% in 2003
and recovered thereafter. End of 2005 the funding ratio was close to 130% again.3 To this end
pension premiums were raised from 7.6% of the wage bill to 16.6%. Employers pay the major
2 The recently introduced collective DC retirement schemes emphasize the collective sharing of inflation, longevity and investment risks and the provision of a satisfactory pension income with a high degree of certainty. 3 Since 1969 pension funds have calculated liabilities using a fixed actuarial discount rate of 4%. In recent years, market valuation of liabilities has started playing a more prominent role. As of 2007 pension funds are required to use the fair value method to calculate their funding ratio. At the end of 2005 the relevant 15-year discount rate was equal to 3.7%.
19
part of these pension premiums. In addition, many employees and pensioners contributed to
the recovery of asset-liability ratios because their pension rights were not fully indexed to
price and wage developments. Note that intergenerational risk sharing through the
compulsory nature of the DB plans protects retirees partly against shocks in asset markets.
The Netherlands does not have an analogue to the PBGC system (safety net).
Second, job change by participants in DB retirement plans in the Netherlands does not
imply a large pension risk. The law requires pension claims to be transferable from one
employer to the other, provided that the pension is adequately funded.4 Third, solvency
requirements make underfunding let alone default a rare phenomenon. More importantly, the
consequences for pension fund participants in case of bankruptcy of their (former) employer
are limited. Because companies and pension funds are separate legal identities, pension claims
are not affected when the company is liquidated. Of course, the flexibility to deal with asset
market developments through intergenerational risk-sharing of active participants and retirees
disappears when the pension fund has to continue without a sponsor.
Summarizing, DB-plans in the Netherlands are relatively safe as far as nominal rights
are concerned, with a low probability of underfunding, and with low bankruptcy or job
change risks. Indexation, however, is conditional upon asset-liability ratios.
3. Methodology and data
3.1 Data and summary statistics
Our data have been collected through an internet-survey among members of the
CentERpanel of CentERdata, a survey research institute that is specialized in internet
surveys.5 The questionnaires are answered at home using an internet connection.6 Thanks to
the internet set-up of the survey, participants do not feel rushed to give an answer and are
fully anonymous when answering the questions.7 Data collected with internet surveys
manifest higher validity and less social desirability response bias than those collected via
telephone interviewing (Chang and Krosnick, 2003). Participants are not paid for their co-
operation. 4 Retirement plans are only one of many conditions in job contracts. Some variation in plans across employers exists. Nevertheless, it is unlikely that employees choose an employer primarily to obtain a specific retirement plan or that employers compete in the labor market through retirement plan conditions. 5 CentERdata forms part of the CentER Group at Tilburg University. See also http://www.uvt.nl/centerdata/nl. 6 Households who do not have access to a pc are provided with a set-top-box for their television 7 In case of attrition of panel-members, CentERdata selects new members to keep the panel representative for the Dutch population High income members are somewhat overrepresented. We have verified that this does not affect the descriptive statistics qualitatively.
20
Our sample consists of those panel members aged 18 and older that are employed, are
looking for a job, or are students. The survey comprises two questionnaires in the period April
2004-January 2005. 1314 respondents (out of 1521) returned the first questionnaire.8 1150
respondents completed the second questionnaire. Our regression analysis of pension attitudes
(Section 4.1 below) is based on the 1134 respondents who completed both questionnaires and
for whom we have information on their education level and monthly income.9 The average
age of the 1134 respondents is 42 years, 58% is male and 91% is employed. The gross
average monthly income of employees is slightly above € 2800 (about € 2300 for the median
employee). We do not use the type of pension scheme as an explanatory variable in our
analysis for a number of reasons. More than 1 out of 3 respondents does not know whether he
has a DB or a DC plan. About 9% state that they have a pure DC retirement plan or a plan
with DC-elements. However, many respondents in this group are covered by collective DC
plans and while being exposed to market risk they have no or a very limited set of choice
options. Further analysis (not reported here) shows that that adding a background variable
indicating current DC coverage does not change our results.
The objective background variables that are used to explain individual behavior are
defined as follows:
Age: respondents’ age measured in years Education: dummy for high education level (1=completed higher vocational or university
education, 0 = other) Single: dummy for being single or having a partner (1=single, 0=married or living together) Male: dummy for gender (1=male, 0=female) Income: respondents’ gross monthly salary in euro
3.2 The first questionnaire
The first questionnaire consists of questions focusing on self-assessment of financial
knowledge, on risk attitude in various domains, on pension plan preferences, and on the
investment of pension wealth in a hypothetical DC scheme. We use the first questionnaire to
8 If this first questionnaire was not completed the first time, we offered the questionnaire for a second and if necessary a third time to the group of non-respondents to improve the response rate (actually some survey weekends fell within typical vacation periods). 9 For consistency, we do not make use of the respondents that did not respond to the second questionnaire (after three trials). Total non-response is fairly randomly distributed on important characteristics as age, income, gender, education and occupation. Selection bias does not seem to be an important problem, though small differences in composition between the sample used and the panel may exist. The descriptive statistics presented in this paper have also been calculated using weights correcting for these differences in composition. Results remain qualitatively unchanged, unless indicated otherwise. All results are available from the authors upon request.
21
construct the following three explanatory variables that are theoretically relevant in individual
choices with respect to retirement schemes:
FinExpert: self-assessment financial expertise (7 classes, from very low to very high) RiskTolSubj: self-assessment risk tolerance (7 classes, from strongly risk averse to risk tolerant) RiskTolObj: theoretical measured risk tolerance (6 classes from strongly risk averse to risk tolerant)
Figure 1 presents the distribution of self-assessed financial expertise and shows that
almost half of the respondents rate their financial expertise as very low (categories 1 and 2).
This is in line with recent literature that typically finds financial knowledge to be quite
limited. Lusardi and Mitchell (2005) for instance, report similar evidence for the US. We
introduce two different measures of risk attitude to account for the unsettled debate in the
literature on the appropriate measurement of risk tolerance. Barsky, Juster, Kimball and
Shapiro (1997) propose a ‘gambles-over-lifetime-income’ approach that is well founded in
economic theory. In particular, they measure risk attitude by offering hypothetical choices
between uncertain labor income streams. However, Kapteyn and Teppa (2002) compare an
extended version of the Barsky et al. approach with simple ad hoc measures of risk attitude
using general questions on risk attitude. Kapteyn and Teppa conclude that the latter perform
better in predicting portfolio choice.10 They do find a positive correlation between the two
measures, though.
10 See also Donkers and van Soest (1999).
Figure 1 Self-assessed financial expertisePercentage of respondents (N=1134)
1 2 3 4 5 6 70
5
10
15
20
25
30
1 = very low7 = very high
22
Table 1 summarizes the evidence on self-assessed risk tolerance. In the survey, we
asked for risk tolerance in general matters, in financial matters and in the pension domain,
respectively. The results show that most respondents consider themselves quite risk averse,
with the level of risk aversion increasing from general issues, to financial issues to pension
issues. The average risk tolerance scores in these categories are 3.2, 2.8 and 2.6 respectively.
Differences are statistically significant at standard levels of significance. In Figure 2, we
present measured risk tolerance as extracted from a life-time income gamble.11 Again, risk
aversion appears high. The correlation between the measures in Table 1 and Figure 2 is
positive but small (0.25). Since they apparently measure different dimensions of risk aversion,
we will use them both in our empirical analysis.
Table 1. Self-assessed risk tolerance in three different domains Percentage of respondents and mean rating ________________________________________________________________________________ Rating risk tolerance General Financial matters Pension domain ________________________________ _____________ _____________ _____________ 1 (try to avoid risks as much as possible) 11.9 16.6 21.3 2 22.0 31.8 31.2 3 24.6 24.5 19.2 4 22.9 15.0 22.4 5 14.9 10.1 4.5 6 3.4 1.9 1.1 7 (like to take a chance) 0.4 0.2 0.3 ________________________________ _____________ _____________ _____________ Mean rating 3.18 2.77 2.62 ________________________________________________________________________________ Note: Total number of observations: 1134. Percentages in columns may not add up to 100 due to rounding. A t-test on the equality of the mean rating for risk tolerance in general and in financial matters is rejected strongly (t=14.7, p-value=0.000). The same applies to the test on the equality of mean ratings of risk tolerance in financial matters and in the pension domain (t=4.5, p-value=0.000).
3.3 The second questionnaire
In the first questionnaire, we asked each respondent to indicate his/her preferred
portfolio investment mix in a hypothetical DC scheme, expressed as the percentage of stocks
and bonds respectively. Based on the given investment mix, we then constructed an
individually tailored future income benefit scheme (in the form of a probability distribution of
the future monthly pension allowance) for each respondent. 11 Respondents are asked to make risky choices in a gamble over lifetime income. In the first round, they must choose between a certain job with fixed income Y, or a job with a 50% chance of an income of 2Y and a 50% chance of an income of aY (a=0.7). In the second round, the choice becomes more or less risky (a equals 0.5 or 0.8) depending on their first choice. Similar to Kapteyn and Teppa (2002), respondents who choose either twice the risky alternative or twice the safe alternative enter a third round (a equals 0.25 or 0.9). Based on their choices, respondents are assigned to one of six categories of different risk appetite.
23
In the second questionnaire, we present each respondent both the probability
distribution of the pension allowance based on one’s own stated preference and the
corresponding probability distribution based on the median investment portfolio. We then ask
the respondent to rate the attractiveness of each portfolio without revealing that one of the two
schemes reflects one’s own investment choice. Section 4.3 explains the details of this
experiment.
4. Empirical results
In this section, we present and analyze the survey results. First, we summarize what
retirement plan participants want and analyze the relation with individual characteristics.
Second, we investigate how respondents claim they would invest their pension savings in the
hypothetical situation of investor autonomy. Third, we assess individual consistency in an
experiment on portfolio choice.
4.1 What do retirement plan participants want?
In order to shed light on the respondents’ attitude towards compulsory retirement
savings, we first asked whether they are happy with a compulsory pension scheme, and if so,
Figure 2 Risk tolerance in a gamble on lifetime incomePercentage of respondents (N=1134)
1 2 3 4 5 60
5
10
15
20
25
30
1 = prefers current lifetime income to a gamble with 50% chance on double the current income and with 50% chance on 90% of the current income.6 = prefers a gamble with 50% chance on double the current income and 50% chance on 25% of the current income above the current lifetime income.
24
for what reason. 12 The majority of respondents (77%) is in favor of compulsory retirement
saving. 12% is against compulsory retirement savings, 6% is indifferent and 5% does not
know.13 About 60% of the respondents in favor of a compulsory pension scheme give the cost
of retirement planning (in terms of time and effort) as the most important reason. One-third
states that otherwise they would not save enough for retirement, indicating awareness of a
self-control problem (see also Thaler and Shefrin, 1981, and Thaler and Benartzi, 2004, for
evidence in support of self-control problems)
Subsequently, we ask for respondents’ preference for a DB or DC type of plan,
respectively. The question explains the trade-off between a DB system with fluctuating
pension premiums in order to guarantee a nominal defined benefit level, versus a DC system
in which premiums are fixed but the benefit varies with investment returns. Almost two-thirds
of the respondents (718 out of 1134 respondents, which is 63 percent of our sample) indicates
a preference for DB over an individual DC system. Only 12 percent prefers a DC system, 10
percent is indifferent and 15 percent doesn’t know. The strong preference for DB confirms
earlier findings in the Netherlands (De Vos, Alessie and Fontein, 1998).
A possible explanation is a status-quo bias (see Samuelson and Zeckhauser, 1988). DB
systems have dominated the Dutch pension landscape for a long time without noticeable
problems. As a result, retirement plan participants are quite satisfied with the current system
and may be reluctant to switch to arrangements that they are unfamiliar with.14 In addition,
financial (and pension-related) literacy in the Netherlands is very limited (Van Els, Van den
End and Van Rooij, 2004). Intuitively, this makes sense. Due to the DB nature of most
pension plans and the compulsory participation in these plans by the large majority of Dutch
employees, incentives to become more financially educated have been low. Guiso, Haliassos
and Jappelli (2002) provide support for the hypothesis that direct exposure to financial
markets, e.g. through individual DC retirement plans, is positively correlated with financial
knowledge and risk tolerance.
A possible additional reason why people might prefer the certainty of the current DB
system is that for many employees in the Netherlands there is virtually no possibility to 12 Response rates differ considerably across questions. Typically the response rates for ‘easy’ (general) questions, such as the self assessment of risk attitudes, are higher than those for intricate questions which require more thought. 13 This is related to the unsettled debate on optimality of life-time savings. See for instance Poterba, Venti and Wise (2004), Lusardi (1999), Gale (1999), Gustman and Steinmeier (1999), Hurd and Rohwedder (2005) and Scholz, Seshadri and Khitatrakun (2004). Our analysis does not provide further direct evidence on this issue. 14 Actually, the wording of the corresponding question in the survey may have contributed to this effect. In the questionnaire, we described the DB system by explaining that it is largely comparable to the current system in the Netherlands. To the extent that this implicitly may have suggested to respondents that the DB systems from the past are sustainable at historical costs, we may have presented the status quo as too inexpensive.
25
postpone the age of retirement – as opposed to for example the US where laws against age
discrimination offer more flexibility on this issue – to compensate for inadequate (DC)
pensions allowances. The official retirement age in the Netherlands is 65. Labor contracts end
automatically when people reach that age. At the time of the survey most collective labor
agreements had even lower official retirement ages (often 60 or 62). As a result, labor
participation rates of the elderly are quite low.
In Table 2, we provide the results of a more in-depth analysis of individual
determinants of pension system preferences using a multinomial probit regression. In panel A
we report the marginal effects of objective personal characteristics on the probability of
choosing between DB and DC. The rows show how a unit change in one of the personal
characteristics affects the probability of an individual choosing one of the answer categories.
T-values are in parentheses. In panel B, three more explanatory variables are included that
represent self-assessed financial expertise and risk tolerance. These latter three variables are
not strictly exogenous as they may be affected by the explanatory variables used in panel A.
A comparison of panels A and B may shed light on this. Importantly, both financial expertise
and risk preferences may be influenced by actual pension plan coverage. Pension plan
participation may affect preferences and thus help people become aware of the need to save
(Cagan 1965; Katona 1965; Gale and Scholz 1994). To account for this, we replicated the
analysis underlying Table 2 including a dummy for coverage in DB or DC plans
respectively.15 The dummy coefficient fails to be significant and results remain virtually
unchanged. Theoretically, it is also possible that preferences affect (the choice of) pension
plan coverage. However, for the Netherlands this is unlikely, given the dominance of and
automatic enrolment in DB plans as well as the fact that pension plan coverage is only one of
many job contract characteristics (see also footnote 4).
A few issues stand out from Table 2. Overall, the explanatory power of the regressions
is limited, but we are unable to reject a test on the joint significance of all regression
coefficients at standard confidence levels. From panel A, we see that respondents in higher
age categories increasingly prefer DB systems over DC systems. The probability of a DB
choice goes up significantly, while that of a DC choice goes down. Also, males choose
significantly more often for DC systems and less often for DB systems than females.
Implicitly, this suggests males are more risk-tolerant on average.16 Higher income as well as
15 We thank an anonymous referee for this suggestion. Unreported results are available from the authors on request. 16 See Jianakoplos and Bernasek (1998) and Schubert, Brachinger, Brown and Gysler (1999) for research on gender and risk tolerance.
26
Table 2. Determinants of the preference for DB and DC A. Marginal effects on probability for each answer; financial expertise and risk attitude excluded __________________________________________________________________________________ Preferences _________________________________________________________________ DB DC Indifferent Don’t know ______________ ______________ ______________ ______________ Age .005 (3.41) -.004 (3.82) .000 (0.22) -.001 (1.40) Log(Income) .016 (1.69) .014 (1.95) -.009 (1.73) -.021 (3.52) Education .037 (1.25) .047 (2.34) -.035 (1.89) -.049 (2.33) Male -.047 (1.56) .075 (3.95) .006 (0.33) -.035 (1.56) Single .044 (1.35) -.006 (0.30) .005 (0.22) -.042 (1.86) B. Marginal effects on probability for each answer; financial expertise and risk attitude included __________________________________________________________________________________ Preferences _________________________________________________________________ DB DC Indifferent Don’t know ______________ ______________ ______________ ______________ Age .003 (2.07) -.002 (1.91) .001 (0.70) -.002 (1.81) Log(Income) .019 (2.07) .009 (1.31) -.010 (1.74) -.018 (3.09) Education .033 (1.11) .023 (1.23) -.028 (1.47) -.028 (1.29) Male -.041 (1.74) .039 (2.10) .012 (0.62) -.011 (0.47) Single .053 (0.84) -.017 (0.91) .001 (0.03) -.036 (1.56) FinExpert .031 (3.06) .020 (3.51) -.018 (2.73) -.033 (4.36) RiskTolSubj -.057 (4.70) .028 (3.90) .028 (3.64) .002 (0.18) RiskTolObj -.028 (2.50) .035 (5.42) .010 (1.39) -.017 (2.01) __________________________________________________________________________________ Note: Number of observations: 1134. Log likelihood: -1156.95 (panel A) and -1093.71 (panel B). Marginal effects are calculated from a multinomial probit regression evaluated at the mean value of explanatory variables (discrete changes from 0 to 1 for dummy variables). Absolute values of t-statistic are in parentheses.
higher educated respondents are less likely to say they are indifferent or don’t know. That is,
they are more inclined to make a choice. This choice may go either way as both the
probability of choosing DB and DC rises, but only the DC effects are (marginally) significant.
With the introduction of three additional indicators in panel B, size and significance of the
explanatory variables from panel A is reduced across the board, suggesting their effect mostly
works indirectly through the impact on financial expertise and risk tolerance measures.17
Instead, financial expertise and risk tolerance enter significantly. Self-assessed financial
expertise reduces the likelihood of ‘don’t know’ answers as well as ‘indifference’ while
increasing the likelihood of preferring DB and DC systems approximately by an equal
amount. Apparently, higher confidence in one’s own financial expertise increases the
likelihood of a clear-cut choice. Higher (self-perceived) risk tolerance increases the likelihood
of choosing a DC system and decreases the likelihood of a DB choice. Interestingly, the risk 17 A further analysis is outside the scope of this paper but will be followed up on in a companion paper. Preliminary regressions indeed show strong dependence of risk tolerance measures on individual characteristics.
27
tolerance indicators appear to be complements rather than substitutes. ‘Age’ remains
significant after controlling for expertise and risk-tolerance. Possibly, the higher preference
for DB versus DC for older people reflects their lower flexibility to compensate for
disappointing pension income (in a DC system) through labor market participation and a
stronger status quo bias due to their longer exposure to a well-functioning DB system.
In a related question, we asked respondents to indicate which percentage of their pre-
retirement net wage income they would want to receive as a guaranteed pension allowance
after retirement. We formulated the survey question in the following way: ‘Imagine a pension
scheme that combines the DB and DC system. Part of the pension benefit is guaranteed
through collective arrangements, but premiums may fluctuate and part of the pension benefit
depends on developments on stock and bond markets but the payable premium is fixed. If you
have to choose a combination of these two systems, what percentage of your net wage income
would you want to be guaranteed as pension benefits? Your answer may vary from 0% to
90%’. Out of the total of 1134 respondents, 870 gave a numerical answer to this question and
264 persons said they didn’t know. The mean preferred percentage was 69 percent, while the
median equaled 70 percent. Only 11 percent of the respondents would be satisfied with a
guaranteed pension income below 50 percent of their net wage income. Hence the willingness
to take risks with future retirement income is very low, which supports the findings on risk
tolerance in the pension domain in Section 3.2.18
In Table 3 we relate the preferred guaranteed pension income to the individual
preference for DB or DC. The results show that choice for a high (low) guaranteed pension
income as a percentage of pre-retirement wage income is strongly positively correlated with
the preference for a DB (DC) system. Out of the 718 proponents of the DB system, 49 percent
(22.0 + 27.0) wants a guaranteed pension income of over 70 percent of wage income. Only 14
percent in this group settles for 50 percent or less. On the other hand, only about 13 percent of
those who prefer a DC system require a certain retirement income in excess of 70 percent,
while over 45% percent in this group is satisfied with an income guarantee of 50 percent or
less. An additional regression to link the preferred guaranteed income percentage to factual
individual characteristics was not very successful. The explanatory power was around 1
percent and only age and gender are individually significant. The preferred guaranteed
retirement income increases with age. Men require on average a 4 percentage points lower
guaranteed pension income than women. Inclusion of financial expertise and risk tolerance
18 Van Els, Van den End and Van Rooij (2004) find that people are willing to pay for the security of guaranteed benefits.
28
indicators significantly improves the explanatory power to over 6 percent. Financial expertise
is insignificant. Both higher objective and higher subjective risk tolerance lower the required
percentage of guaranteed income.19
Table 3. Preferred income guarantee versus preferred pension system Percentages of respondents preferring the specified pension system ________________________________________________________________________________ Preferred net pension income guarantee as a percentage of net final wage income
________________________________________________________________ Preferred system (# respondents) <= 50 >50 and <=60 >60 and <=70 >70 and <=80 >80 and <=90 DK _____________ _____ ___________ ___________ ___________ ___________ ____ DB (718) 13.5 7.2 19.8 22.0 27.0 10.5 DC (136) 45.6 15.4 19.1 8.8 3.7 7.4 Indifferent (115) 27.8 5.2 13.0 17.4 7.8 28.7 DK (165) 2.4 1.2 1.2 1.8 4.9 88.5 ________________________________________________________________________________ Note: DK = ‘Do not know’; Percentages in rows may not add up to 100 due to rounding.
The caveats with respect to the interpretation of the results on the DB-DC preference
question apply again. A current lack of financial expertise, unfamiliarity with DC-elements
and a high level of risk aversion may be partly the result of the prevailing compulsory DB
schemes for most respondents that do not provide direct exposure to financial decision
making and consequently do not provide many incentives to become more educated. This
may contribute to the attractiveness of the status quo situation. In most Dutch retirement plans
the implicit rule of thumb (status quo) is that after retirement gross income can be expected to
equal about 70 percent of final gross wage, which after tax would correspond to about 90
percent of pre-retirement net income due to additional tax advantages of the retired.
Nevertheless, Table 3 shows that there is a lot of heterogeneity in the stated preferences. Only
a minority would opt for a full DB retirement plan. Overall, the main message is that the
(un)willingness to accept uncertainty with respect to retirement benefits strongly positively
correlates with a preference for a DC (DB) system. In that sense, the results are internally
consistent.
To assess the desired degree of autonomy in portfolio investments, we have asked
whether, in the hypothetical situation of an individual DC scheme, respondents would want to
have control over individualized pension fund accounts (having the opportunity to invest their
pension money according to an investment profile offered by their pension fund or according
to their own investment choices) or whether they would delegate this to the pension fund.
19 Detailed regression results are available from the authors upon request.
29
Almost half of our sample (48.4%) would leave investment decisions to the pension fund,
26.1% prefers autonomy, 10.6% is indifferent and 14.9% doesn’t know. Table 4 summarizes
the link with pension system preferences. The set-up is analogous to Table 3. Again, the
dichotomy between supporters of DB versus DC systems is clear-cut. A large majority of the
former group prefers the pension fund to decide on investments, while an equally large
majority of the latter group prefers individual autonomy.
Table 4. Preference for investor autonomy versus preferred pension system Percentages of respondents preferring the specified pension system _____________________________________________________________________ Preferred system (# respondents)
Pension fund Investor autonomy
Indifference Don’t know
_______________ ___________ ___________ ___________ ___________ DB (718) 60.7 24.5 9.1 5.7 DC (136) 33.8 56.6 7.4 2.2 Indifferent (115) 37.4 23.5 30.4 8.7 Don’t know (165) 14.6 9.7 6.1 69.7 _____________________________________________________________________ Note: Percentages in rows may not add up to 100 due to rounding.
The results of a multinomial probit regression of the preferred degree of investor
autonomy on individual characteristics are presented in Table 5. The setup is similar to Table
2. In Panel A most coefficients are insignificant. Only the probability of respondents
preferring delegation to the pension fund significantly increases with education, while men
are significantly more likely to choose investor autonomy. When the indicators for financial
expertise and risk tolerance are included (panel B), the education effect remains significant,
while the gender effect disappears. Self-assessed higher financial expertise is shown to
strongly increase the preference for individual autonomy and to marginally reduce the
preference for pension fund control. The risk tolerance indicator based on the Barsky life-time
income gamble has a strong positive impact on the probability of investor autonomy. This
finding opposes the conclusion by Kapteyn and Teppa (2002). Typically, and in line with
intuition, income, education, financial expertise and risk tolerance all reduce the likelihood
that respondents answer ‘don’t know’ in panels A and B.20
In order to see whether an increase in financial expertise would change the preference
for investor autonomy we have asked respondents whether the opportunity to take a course
(for free) to upgrade their financial expertise would affect their willingness to take control
over their retirement savings. 42% respond that financial education would make them more
20 When included the dummy on current DC coverage is insignificant and does not change the results.
30
inclined to take control of their retirement portfolio; another 42% believes it would not, and
the remaining 16% do not know. The answers are correlated with the respondents’ preference
for DC over DB: six out of ten respondents who prefer a DC system believe that they would
take more control over retirement savings investment when offered the possibility to upgrade
their financial knowledge.
Table 5. Determinants of the preferred degree of autonomy A. Marginal effects on probability for each preference excluding financial expertise and risk attitude ________________________________________________________________________________ Preferences ________________________________________________________________ Pension fund Investor
autonomy Indifferent Don’t know
______________ _______________ ______________ ______________ Age .002 (1.12) .000 (0.32) -.001 (0.62) -.002 (1.44) Log(Income) .015 (1.55) .004 (0.51) -.005 (0.89) -.015 (2.48) Education .064 (2.08) .027 (0.98) -.007 (0.39) -.084 (4.01) Male -.038 (1.22) .083 (3.07) -.025 (1.27) -.020 (0.89) Single -.033 (0.96) .040 (1.28) -.020 (0.99) .013 (0.54) B. Marginal effects on probability for each preference including financial expertise and risk attitude ________________________________________________________________________________ Preferences ________________________________________________________________ Pension fund Investor
autonomy Indifferent Don’t know
______________ _______________ ______________ ______________ Age .001 (0.56) .002 (1.39) -.001 (0.54) -.002 (2.14) Log(Income) .018 (1.87) -.003 (0.34) -.005 (0.79) -.011 (1.86) Education .078 (2.46) -.016 (0.56) .001 (0.03) -.063 (2.95) Male -.017 (0.51) .022 (0.78) -.015 (0.73) .009 (0.43) Single -.029 (0.84) .032 (1.01) -.023 (1.09) .021 (0.82) FinExpert -.016 (1.51) .067 (7.37) -.017 (2.46) -.034 (4.58) RiskTolSubj -.009 (0.72) .009 (0.77) .009 (1.17) -.009 (0.99) RiskTolObj -.015 (1.28) .034 (3.26) .004 (0.52) -.022 (2.72) ________________________________________________________________________________ Note: Number of observations: 1134. Log likelihood: -1360.16 (panel A) and -1307.75 (panel B). Marginal effects are calculated from a multinomial probit regression evaluated at the mean value of explanatory variables (discrete changes from 0 to 1 for dummy variables). Absolute values of t-statistic in parentheses.
In summary, we firstly conclude that a large majority of our sample prefers a DB
system over a DC system, prefers a relatively high percentage of current income to be
guaranteed after retirement, and prefers a professional pension fund to decide about portfolio
investment for retirement. Moreover, respondents generally make internally consistent
choices on these different items. Secondly, respondents are quite risk averse on average,
especially in the pension domain, and financially illiterate. We find that that self-assessed and
31
measured risk attitudes as well as self-assessed financial expertise are significant explanatory
variables with respect to pension scheme preferences. Respondents who are more inclined to
take risk and consider themselves to be financially sophisticated, are more likely to prefer a
DC plan, are more likely to prefer a relatively low guaranteed retirement income, and are
more likely to prefer investor autonomy. The effects of other explanatory variables like age,
gender, income, or education on pension preferences are mostly small and insignificant once
financial expertise and risk tolerance are accounted for. An important caveat in our analysis is
the potential endogeneity of (self-assessed) risk tolerance and financial literacy as these
themselves may be functions of exposure to a specific pension scheme.
4.2 What would retirement plan participants do in case of investor autonomy?
We now focus on the issue how pension plan participants (say they) would behave if
they would have to make their own investment decisions.21 In the hypothetical situation of
individualized DC pension accounts we first ask how much of the retirement savings portfolio
a respondent would invest in stocks and bonds respectively. The preferred percentage of
stocks in the portfolio ranges from zero (portfolio completely in the form of bonds) to 100
(portfolio completely in stocks) percent. Figure 3 presents the results. Out of 1134
respondents, 877 give a numerical answer. Only those respondents are included in Figure 3.
21 In this respect it is worthwhile to stress that empirical evidence indicates that individuals are highly sensitive to defaults, notably in the area of investing for retirement (Cronqvist and Thaler, 2004)
Figure 3 Preferred retirement savings composition: percentage stock in individual portfolioPercentage of respondents (N=877)
0 0<10 10<20 20<30 30<40 40<50 50<60 60<70 70<80 80<90 90<100
Percentage of stock investments
0
5
10
15
20
25
30
32
The median and mean response both equal 30 percent. Those who previously indicated to
prefer a DC system (128 out of 877) on average apparently prefer to hold a larger part of their
portfolio in stocks than those who prefer a defined benefit system (610 respondents). The
mean response of the former group equals 39.6 percent, while that of the latter equals 28.7
percent. The preferred percentage of stock by the respondents in our sample is below that
typically found in US DC schemes where participants choose the composition of their
retirement savings portfolio (Benartzi and Thaler, 2002). Poterba, Venti and Wise (2003)
conclude that households that do not have extremely high risk-aversion would be better of, ex
ante, by holding a portfolio of stocks rather than bonds. Potential explanations for the
relatively low preferred percentage of stocks are the timing of the survey in 2004 – with the
stock market decline over the period 2000-2003 fresh in mind – as well as low financial
expertise and/or risk tolerance.
In Table 6, we display results for two related regressions with the preferred stock
percentage as the dependent variable. In the first regression, only strictly exogenous personal
characteristics are used as explanatory variables. The results only show a significant gender
effect. Being male (female) increases (decreases) the proportion of stock holdings by 6
percentage points on average. In the second regression, indicators of financial expertise and
risk attitude are added. Overall, the explanatory power substantially increases. Financial
expertise and the willingness to take risks appear the most important factors in determining
Table 6. Determinants of preferred retirement portfolio composition __________________________________________________________ Preferred percentage of stocks ______________________________________ A B __________________ _________________ Age -.075 (1.11) .056 (0.86) Log(Income) .367 (0.87) .057 (0.12) Education 1.635 (1.15) -.070 (0.05) Male 6.051 (4.11) 2.590 (1.82) Single .304 (0.19) -.121 (0.08) FinExpert 2.290 (5.13) RiskTolSubj 3.418 (6.21) RiskTolObj 2.237 (4.25) Constant 26.171 (6.49) 3.961 (0.92) Adj R2 0.018 0.141 __________________________________________________________ Note: Number of observations: 877. Regression coefficients from an ordinary least squares regression. Regression A (B) excludes (includes) financial expertise and risk attitude. Absolute values of t-statistic in parentheses.
33
the preferred portfolio mix. The gender effect is reduced in size but remains marginally
significant. The finding that woman are more risk averse is consistent with earlier research
(Jianakoplos and Bernasek 1998), although other studies come to opposite conclusions (e.g.
Schubert, Brachinger, Brown and Gysler, 1999). Our results imply that even after controlling
for risk tolerance, women prefer less stock in their retirement savings portfolio.
Following the survey question on the preferred stock percentage, we first asked which
factors played a role in that (initial) choice, and second which factors could play a role in the
future in adjusting the preferred percentage of stocks in the total portfolio. Table 7
summarizes the outcomes. Overall, about ninety percent of the respondents states that either
personal circumstances – age, family composition, personal financial position or accumulated
pension claims – or general economic conditions and financial markets expectations have
played a role in choosing the preferred portfolio mix and will continue to influence future
adjustment decisions. About 10 percent regards none of these factors as relevant for either the
initial portfolio composition or future adjustments. Apart from age, determinants of initial
choice of portfolio and determinants of later changes are more or less the same. Interestingly,
44 percent of the respondents indicate age as a relevant determinant of their initial choice,
whereas our regression analysis (see Table 6 above) indicates that in a multivariate analysis,
age is not significant. The percentage of respondents that believes age to be an important
factor for future portfolio adjustments is much lower, namely 23 as opposed to 44 for the
initial composition. An explanation might be that age – unlike the other factors – is perfectly
predictable. People may choose their optimal portfolio mix now given their (known) time to
retirement and may not consider future changes. All other factors are subject to (unexpected)
changes and as such may require portfolio adjustments from the perspective of the
respondents. Not all factors are equally important, though. Especially one’s personal financial
situation (49 percent) and the two indicators of general economic circumstances (59 and 46
percent, respectively) apparently are strong drivers of future investment decisions. About half
of our respondents would consider changing the own portfolio mix in case of (important)
changes in any of these three variables. Assuming that changes in these circumstances are
quite likely to happen more than once over, say, a ten-year period, this implies a relatively
high (stated) degree of activism in portfolio management. Interestingly, empirical evidence in
previous studies (for example Ameriks and Zeldes, 2001) typically reports a much lower
degree of activism. Possibly, the respondents in our sample overestimate their activism.
34
Table 7. Underlying determinants of initial choice and subsequent changes in retirement portfolio composition according to respondents Percentage of respondents naming the category ___________________________________________________________________ Determinant Initial composition Reasons for change __________________________ __________________ __________________ None 9.6 8.3 Age 44.4 23.4 Family composition 24.3 18.1 Personal financial situation 52.9 49.4 Accumulated pension wealth 38.8 31.9 General economic condition 49.8 59.3 Financial market expectations 39.9 45.9 Other 1.9 1.6 ___________________________________________________________________
Note: Number of observations: 877. Percentages do not add up to 100 as respondents could give several answers.
Subsequently, we relate the stated degree of activism to individual characteristics
using probit regressions with the dependent variable being 1 when respondents state that
specific individual or general circumstances do not influence their choice of portfolio mix and
0 when they state that one or more of the arguments in Table 7 play a role in their decision. In
Table 8, we report the results. Generally, the overall explanatory power is low and
coefficients are insignificant. Only self-assessed financial expertise has a sizable and
significant effect on the degree of stated activism. Apparently, higher perceived financial
expertise leads individuals to stronger believe in the information in specific economic
indicators and/or in their own ability to interpret such information. Thus, increased financial
expertise may be an important driving force for people’s willingness to actively manage their
portfolio.22 Risk tolerance, on the other hand, does not play any role.
In summary, when forced to choose, respondents pick a relatively safe portfolio with
only about 30 percent of stocks on average. High risk tolerance and self-assessed expertise –
as well as being male – raise the chosen percentage of stocks in portfolio. Respondents also
think they will be quite activist in changing their portfolio when conditions change. Especially
those who think they are better experts are inclined to (say they will) change their portfolio.
22 Similar regressions to those in Table 8 were performed for individual drivers of portfolio choice and portfolio adjustment respectively. That is, the dependent variable was set to one when a respondent stated that a specific determinant - say one’s personal financial position - would not be a factor in his portfolio choice or adjustment and to zero when it would. Consistently, higher financial expertise is significant with a negative sign in these regressions. The effect is strongest for financial market expectations, family composition and age, and weakest for the case of general economic conditions and one’s personal financial situation.
35
Table 8. Determinants of planned activism ___________________________________________________________________ Initial composition does not
depend on economic factors Change in composition does not
depend on economic factors _________________________ _________________________ Age -.000 (0.08) -.000 (0.49) .000 (0.27) -.000 (0.19) Log(Income) -.008 (1.27) -.006 (1.00) -.011 (2.05) -.010 (1.80) Education -.027 (1.32) -.014 (0.70) -.030 (1.56) -.019 (1.03) Male -.008 (0.38) .013 (0.62) -.002 (0.13) .016 (0.84) Single -.018 (0.80) -.016 (0.74) -.018 (0.88) -.016 (0.81) FinExpert -.027 (3.84) -.021 (3.18) RiskTolSubj .002 (0.30) -.002 (0.28) RiskTolObj -.006 (0.84) -.007 (0.93) Loglikelihood -265.21 -274.32 -246.99 -239.65 Pseudo R2 0.009 0.033 0.017 0.047 ___________________________________________________________________
Note: Number of observations: 877. Marginal effects on probability are calculated from probit regressions evaluated at the mean value of explanatory variables (discrete changes from 0 to 1 for dummy variables). Absolute values of t-statistic in parentheses. 4.3 Are retirement plan participants able to choose?
In order to further investigate whether respondents are able to make retirement
investment choices and have well-defined preferences, we carry out an experiment similar to
the one by Benartzi and Thaler (2002). We asked respondents in our sample to rate the
attractiveness of two benefit schemes on a 5-points scale. One scheme was based on each
individual’s own preferred portfolio (as answered in the first questionnaire), the other on the
median portfolio choice of all respondents (consisting of 30 percent stocks and 70 percent
bonds). We did not reveal that one of the two benefit schemes was based on the individually
chosen portfolio.
We constructed the benefit schemes using Monte Carlo simulations. The distribution
of retirement income conditional on gross salary of the respondent, the percentage of stock
investment, and the mean and volatility of bond and stock returns is determined using 1000
runs with a 40 year horizon. We assume annual bonds returns to be drawn randomly from a
distribution with a mean of 5 percent and a standard deviation of 4 percent, taking account of
persistence in interest rates. Stock returns are randomly drawn from a distribution with a
mean of 8 percent and a standard deviation of 18 percent. These distributional assumptions
match historical data and are quite standard in pension funds’ ALM analysis. We assume the
annual premium contribution to be 13 percent of the individual’s gross wage, which currently
is the break-even point in a typical Dutch pension scheme (see Van Rooij, Siegmann and
Vlaar, 2004). The premium contributions are invested in stocks and bonds (with an average
36
maturity of 5 years) according to the portfolio mix chosen by the respondent. The portfolio is
rebalanced at the end of each year as to maintain the chosen mix of stock and bond
investment. After each 40-year run, final wealth is invested in an annuity, the annual pension
benefit of which is assumed to equal 1/15th of final wealth (roughly based on Dutch mortality
tables). We then confront each respondent with the 5th, 50th and 95th percentile of the resulting
individualized benefit scheme.
Box 1 presents an example for the case of an employee with a gross salary of € 2300
per month who indicated his preferred portfolio consisted of equal proportions stocks and
bonds. Pension scheme I is the result of an investment strategy of 30% stocks and 70% bonds
(the median choice) and pension scheme II is the result of the investor’s own preferred
portfolio. The latter one has a higher upward potential but also more risk on the downside.
Note though that the pattern is asymmetric. The extra downside risk of portfolio II is
relatively small compared to its extra upside potential. The numbers exemplify 1) the fact that
the riskiness in term of holding stocks is relatively smaller - at least in terms of probabilities -
on a long horizon than on a short one and 2) the fact that yearly contributions become
relatively more (less) important when accumulated wealth is eroded (increased) by a number
of years with disappointing (encouraging) returns in the stock market.
Box 1 Question on the rating of pension schemes
Consider two pension schemes without guaranteed pension benefit. The actual benefit depends on among others general economic and financial market developments. The table below presents the retirement benefit you may expect under each benefit scheme. The numbers present gross benefits in euro per month. The pension contributions you have to pay are equal in both arrangements. The numbers are excluding the state pension benefit (this gross benefit equals € 921 per month for singles and € 632 per person each month for married couples and people living together). Any pension plan you may have arranged on top of these arrangements, is not included. For each benefit scheme we present three possible outcomes (an unfavorable scenario, a favorable scenario and a middle variant).
Economic scenario Pension scheme I Pension scheme II VERY UNFAVOURABLE 610 540 AVERAGE 920 1012 VERY FAVOURABLE 1414 1920
The interpretation of these numbers is as follows. There is a 5% probability of a retirement income above the VERY FAVOURABLE retirement income, there is a 50% probability of a retirement income above the AVERAGE retirement income and there is a 5% probability of a retirement income below the VERY UNFAVOURABLE retirement income. QUESTION: How do you rate these two pension schemes on a scale from 1 to 5 with 5 being ‘very attractive’ and 1 being ‘very unattractive’?
37
In Table 9, we present the resulting distribution of ratings. We distinguish between
three groups of respondents: those who prefer a relatively safe portfolio (334 respondents
with a preferred percentage of stocks less than or equal to 20), those who prefer a relatively
risky portfolio (289 respondents with a preferred percentage of stock greater than or equal to
40), and those with an average portfolio (254 respondents with a preferred stock percentage
between 20 and 40 percent). The first two groups unknowingly rate their own portfolio and
the median portfolio, the last one its own portfolio with the 50 percent portfolio.23
Table 9. Attractiveness of two investment portfolios for three groups of respondents Percentage of respondents and mean rating ___________________________________________________________________________________
%stocks<=20% 20<%stocks<40 %stocks>=40 __________________ ______________________ __________________
Rating attractiveness portfolio Median (30%) Own Median (30%) 50 percent Median (30%) Own _________________ ___________ _____ ___________ ________ ___________ _____ 1 (very unattractive) 4.5 10.2 3.9 2.8 4.8 1.7 2 7.8 21.9 19.3 10.6 19.0 15.2 3 (neutral) 30.2 47.3 47.6 31.1 46.7 27.0 4 42.2 17.7 26.8 45.3 24.9 39.5 5 (very attractive) 15.3 3.0 2.4 10.2 4.5 16.6 Mean rating 3.56 2.81 3.04 3.50 3.05 3.54 T-test (H0: mean ratings are equal)
T=10.29 (p=0.000) T=5.78 (p=0.000) T=6.12 (p=0.000)
___________________________________________________________________________________ Note: Total number of observations: 877 (334 respondents indicated to invest 20% stock or less, 254 respondents indicated to invest between more than 20% but less than 40% stocks and 289 indicated to invest 40% stocks or more). Percentages in columns may not add up to 100 due to rounding.
As Table 9 shows, across all groups, respondents prefer the most risky of the two
portfolios presented to them, regardless of their initial choice. Consider first the group of risk
averse respondents (<20% stocks). On average, they rate the pension scheme based on the
median portfolio 3.56 as opposed to 2.81 for the pension scheme based on their own preferred
portfolio. In fact, 61.1 percent of this group of conservative investors prefers the median to
the own portfolio, while only 15.6% prefers the own portfolio and 23.3% of the group is
indifferent. Put differently, ex post they regret their original choice. The group of respondents
that favored a risky portfolio (>40% stocks), on the other hand rates the median portfolio as
less attractive than their own (average ratings of 3.05 and 3.54 respectively) on average and
typically sticks to the original choice. 59.5 percent of the respondents in this group prefer the
own portfolio to the median. Finally, the middle group that chose a preferred stock percentage
23 In fact, for simplicity we approximate individually chosen own portfolios with stock percentages between 20 and 40 percent with the uniform mean portfolio, which has 30 percent stocks.
38
close to the median (between 20 and 40 percent) rates the own portfolio on average at 3.04
and the more risky 50 percent portfolio at 3.50. 55.9 percent of this group prefers the 50
percent portfolio to their own original choice.24
Our results contrast markedly with Benartzi and Thaler (2002). Whereas in their
sample respondents tend to go for the median portfolio irrespective of their initial portfolio
choice, our respondents unambiguously choose the riskier portfolio of the two, regardless of
their initial choice. Other empirical evidence on the consistency of preferences is mixed as
well, see Benartzi and Thaler (2001) and Huberman and Jiang (2004).
One explanation for our finding – see Benartzi and Thaler (1999) – is that our
respondents are subject to ‘myopic loss aversion’. Under this assumption, the probability of a
short-term loss receives too much weight in long-term portfolio decisions. When confronted
with the (true) distribution of long-run returns, individuals then switch to more risky
portfolios. A second explanation could be that the respondents’ expectations of asset markets
may differ from ours. The observed shift towards a preference for more risky portfolios then
arises from the fact that respondents are more pessimistic on the stock market’s risk return
profile relative to the bond market than implied by our assumptions underlying the Monte
Carlo simulations.25 Note that at the time of the survey in 2004, stock market returns had been
disappointing for a number of years. Either way, both explanations suggest that many
respondents currently lack the skills for being in charge of the investment for retirement
purposes. The shift towards a more risky portfolio by a majority of the respondents suggests
that either the initial preferences were fragile and not firmly grounded or that they were based
on an unrealistically pessimistic – in comparison with historical averages – assessment of the
distribution of stock versus bond returns.26
5. Summary and conclusions
In the Netherlands, the large majority of employees compulsory participates in the DB
retirement plan of their employer. For a long time, the national pension system worked
24 We have also replicated the analysis in Table 9 separately for those respondents covered by DC and DB plans. The results do not show significant differences. In interpreting all these results it is important to note that the sample with DC-participants is small and heterogeneous, including those participating in collective retirement plans without choice options. 25 Potentially, the results may be affected by the emergence of important new economic information in the period between the two surveys that may have changed respondents’ view on the attractiveness of the various portfolios. Given the lack of major economic developments during the period under investigation, we doubt the quantitative importance of such effect. 26 One might argue that respondents are sensitive to the way the distribution of pension benefits is presented. However, Benartzi and Thaler (1999) show that individuals are not very sensitive to variations in the presentation of the retirement income distribution.
39
satisfactorily. The average employee typically never had much influence on pension fund
policies and – possibly as a consequence – is relatively ignorant about many of his own
retirement plan’s details. Although employee confidence in the system and in the safety of
future pension allowances has been and still is very high, the fall in stock prices around the
turn of the century, the new international accounting standards and the ageing of the
population have triggered a debate on the sustainability and design of the system. Many funds
have switched from a system where pension benefits are based on the final gross wage to a
system that is based on career-average wages. After the 2000-2003 stock market crash,
pension premiums were raised while indexation was cut by many pension funds to
compensate for large investment losses. Some pension funds in the Netherlands have started
to experiment with mixed DB/DC and even full (collective) DC systems.
Against this background, this paper provides evidence on pension preferences and
investor autonomy with respect to retirement savings in the Netherlands. The focus is on
whether employees are willing and able to deal with more retirement plan choice. We use
questionnaire responses from about 1000 members of the household panel run by
CentERdata, which is a representative sample of the Dutch population.
Our main conclusions are the following. Risk aversion is quite high on average. When
asked about risk attitude with respect to general matters, financial matters and pension matters
respectively, risk aversion is highest in the pension domain. Simultaneously, the typical
employee considers himself to be financially illiterate. Lack of exposure to self-directed
savings plans and investments may go some way in explaining both the low level of self-
assessed financial expertise and the high level of self-assessed risk aversion. However, US
evidence indicates that financial literacy has not disappeared with the widespread introduction
of individual DC plans (Lusardi and Mitchell, 2005).
The vast majority of Dutch employees is in favor of compulsory saving for retirement
and favors a DB pension system. This preference is, according to the employees, primarily
due to their wish not to spend time on retirement savings decisions and to a perceived self-
control problem, but respondents may also be affected by a status quo bias. If offered a
combined DB/DC system, the majority would choose a guaranteed pension income of 70% or
more of their net labor income. In case of the introduction of an individual DC scheme, most
respondents would prefer to delegate decisions about the investment of their retirement
portfolio to a pension fund. The possibility to enroll in a program for free to improve financial
expertise would only induce a minority - 42 percent - of the employees to become more
supportive of investor autonomy.
40
Self-assessed risk tolerance and financial expertise are important explanatory variables
of pension system attitude. Respondents who are more inclined to take risk and consider
themselves to be financially sophisticated, are more likely to prefer a DC plan, to prefer a
relatively low guaranteed retirement income, and to prefer investor autonomy. When given
investor autonomy, the typical respondent chooses a conservative portfolio with stocks
making up only 30 percent of the average portfolio. Employees expect to be quite activist in
managing the composition of their retirement savings portfolio if they were forced to investor
autonomy in an individual DC scheme. Drivers behind a planned change in portfolio
composition are changes in one’s personal financial situation, general economic conditions,
and expectations of financial markets. Respondents with more confidence in their financial
expertise and lower risk aversion choose more risky portfolios and are more inclined to
actively manage their portfolio when circumstances change.
In a final experiment we show that respondents who originally stated that they would
prefer a relatively safe investment portfolio for retirement with only a small proportion of
stocks tend to switch to a riskier portfolio when shown the distribution of long-run returns on
their own portfolio and the mean risk portfolio (containing more stocks). The same holds for
investors who originally opted for a mean risk portfolio. They too switch too a more risky
portfolio. This result suggests that many Dutch pension plan participants currently lack the
necessary skills to be in charge of their own investment portfolio for retirement purposes.
Our finding that most respondents are reluctant to switch from a DB to a DC system
with more freedom of choice is not surprising, given their high risk aversion in the pension
domain and their low self-assessed degree of financial expertise. However, risk tolerance and
financial literacy are likely to be endogenous, possibly depending on individuals’ exposure to
self-directed savings and investment plans. The conclusions in this paper are conditional on
current preferences and knowledge. Changes in the institutional design may change
preferences and behavior. Be that as it may, changes in the pension scheme design will affect
financial expertise and risk attitude of employees only very gradually. The policy implication
therefore is, that in case of a change over to more individualized DC plans in the Netherlands,
many employees would have to be guided in their retirement planning, for example by
mandatory collective arrangements, made-to-measure defaults or plans offering the possibility
of commitment to long-term savings strategies. Even within DC systems, there are important
gains to be had from maintaining compulsory savings to avoid the pitfall of undersaving due
to self-control problems and procrastination. Our results indicate that a large fraction of
employees is aware of this problem. Similarly, given myopic loss aversion and lack of
41
financial expertise, a switch to a system with full autonomy over investment choices is far
from recommendable. In any case more freedom of choice in employee retirement plans
would have to be accompanied by appropriate default portfolios and/or a limited menu of
investment options with differing risk characteristics. Finally, liberalization of the market for
pension investment would increase the importance of supervision and regulation of the market
for professional pension advice.
42
References Ameriks, J., and S. Zeldes, 2001, How do household portfolio shares vary with age?, mimeo, Columbia Graduate School of Business. Barsky, R., F. Juster, M. Kimball, and M. Shapiro, 1997, Preference parameters and behavioral heterogeneity: An experimental approach in the Health and Retirement Study, Quarterly Journal of Economics, 112, 537-579. Benartzi, S., and R. Thaler, 1999, Risk aversion or myopia? Choices in repeated gambles and retirement investments, Management Science, 45, 364-381. Benartzi, S., and R. Thaler, 2001, Naïve diversification strategies in defined contribution saving plans, American Economic Review, 91, 79-98. Benartzi, S., and R. Thaler, 2002, How much is investor autonomy worth?, 2002, Journal of Finance, 57, 1593-1616. Cagan, P., 1965, The Effect of Pension Plans on Aggregate Saving: Evidence from a Sample Survey, NBER, New York. Chang, L., and J. Krosnick, 2003, Comparing oral interviewing with self-administered computerized questionnaires: an experiment, mimeo, Ohio State University. Choi, J., D. Laibson, B. Madrian, and A. Metrick, 2004, For better or for worse: default effects and 401(k) savings behaviour. In: Z. Bodie, J. Shoven, and D. Wise (eds.), Perspectives on the Economics of Aging, University of Chicago Press, Chicago. Cronqvist, H, and R. Thaler, 2004, Design choices in privatized social-security systems: learning from the Swedish experience, American Economic Review, 94, 424-428. De Vos, K., R. Alessie, and P. Fontein, 1998, Pensioenpreferenties (Pension preferences), Economisch Statistische Berichten, 83, 398-399. Donkers, B, and A. van Soest, 1999, Subjective measures of household preferences and financial decisions, Journal of Economic Psychology, 20, 613-642. Gale, W., 1999, Comments to Annamaria Lusardi, Information, expectations and savings for retirement. In: H. Aaron (ed.), Behavioral Dimensions of Retirement Economics, Brookings Institutions Press, Washington DC/Russel Sage foundation Newark, New York, 116-123. Gale, W., and J. Scholz, 1994, IRAS and household saving, American Economic Review, 84, 1233-1260. Guiso, L., M. Haliassos, and T. Jappelli, 2002, Household Portfolios, MIT Press, Cambridge. Gustman, A., and T. Steinmeier, 1999, Effects of pensions on savings: analysis with data from the Health and Retirement Study, Carnegie Rochester Conference Series on Public Policy, 50, 271-324.
43
Huberman, G., and W. Jiang, 2004, The 1/N heuristic in 401(k) plans, Columbia Business School Working Paper. Hurd, M., and S. Rohwedder, 2005, Changes in consumption and activities at retirement, DNB Working Paper, 39. Jianakoplos, N., and A. Bernasek, 1998, Are women more risk averse? Economic Inquiry, 36, 620-630. Kapteyn, A., and F. Teppa, 2002, Subjective measures of risk aversion and portfolio choice, mimeo, RAND. Katona, G., 1965, Private Pensions and Individual Saving, University of Michigan. Lusardi, A., 1999, Information, expectations, and savings for retirement. In: H. Aaron (ed.), Behavioral Dimensions of Retirement Economics, Brookings Institutions Press, Washington DC/Russel Sage foundation Newark, New York. Lusardi, A., and O. Mitchell, 2005, Financial literacy and planning: implications for retirement wellbeing, DNB Working Paper, 78. Mitchell, O., and S. Zeldes, 1996, Social security privatization: a structure for analysis, American Economic Review, 86, 363-367. Poterba, J., S. Venti, and D. Wise, 2003, Utility evaluation of risk in retirement saving accounts, NBER Working paper, 9892. Poterba, J., S. Venti, and D. Wise, 2004, The transition to personal accounts and increasing retirement wealth: macro- and microevidence. In: D. Wise (ed.), Perspectives on the Economics of Aging, University of Chicago Press. Samuelson, W., and R. Zeckhauser, 1988, Status quo bias in decision making, Journal of Risk and Uncertainty, 1, 7-59. Scholz, J., A. Seshadri, and S. Khitatrakun, 2004, Are Americans saving “optimally” for retirement?, mimeo, University of Wisconsin. Schubert, R., H. Brachinger, M. Brown, and M. Gysler, 1999, Are women really more risk-averse? American Economic Review, 89, 381-385. Thaler, R., and H. Shefrin, 1981, An economic theory of self-control, Journal of Political Economy, 89, 392-406. Thaler, R., and S. Benartzi, 2004, Save more tomorrow: using behavioural economics to increase employee saving, Journal of Political Economy, 112, 164-187. Van Els, P., W. van den End, and M. van Rooij, 2004, Pensions and public opinion: a survey among Dutch households, De Economist, 152, 101-116.
44
Van Rooij, M., A. Siegmann, and P. Vlaar, 2004, PALMNET: A pension asset and liability model for the Netherlands, DNB Research Memorandum WO, 760.
45
PAPER TWO *
FINANCIAL LITERACY AND STOCK MARKET PARTICIPATION
(Co-authors: Annamaria Lusardi and Rob Alessie)
* Basically this is the October 2007 version that appeared in the NBER working paper series with the incorporation of a number of changes upon suggestions by the members of the reading committee. In the third and fourth paper of this thesis, it is referred to as Van Rooij, M., A. Lusardi, and R. Alessie, 2007, Financial literacy and stock market participation, NBER Working Paper, 13565.
47
Financial Literacy and Stock Market Participation
Maarten van Rooij (De Nederlandsche Bank and Netspar) Annamaria Lusardi (Dartmouth College and NBER)
Rob Alessie (University of Groningen, Netspar and Tinbergen Institute)*
November 2008
Abstract Individuals are increasingly put in charge of their financial security after retirement. Moreover, the supply of complex financial products has increased considerably over the years. However, we still have little or no information about whether individuals have the financial knowledge and skills to navigate this new financial environment. To better understand financial literacy and its relation to financial decision-making, we have devised two special modules for the DNB Household Survey. We have designed questions to measure numeracy and basic knowledge related to the working of inflation and interest rates, as well as questions to measure more advanced financial knowledge related to financial market instruments (stocks, bonds, and mutual funds). We evaluate the importance of financial literacy by studying its relation to the stock market: Are more financially knowledgeable individuals more likely to hold stocks? To assess the direction of causality, we make use of questions measuring financial knowledge before investing in the stock market. We find that, while the understanding of basic economic concepts related to inflation and interest rate compounding is far from perfect, it outperforms the limited knowledge of stocks and bonds, the concept of risk diversification, and the working of financial markets. We also find that the measurement of financial literacy is very sensitive to the wording of survey questions. This provides additional evidence for limited financial knowledge. Finally, we report evidence of an independent effect of financial literacy on stock market participation: Those who have low financial literacy are significantly less likely to invest in stocks. Key words: Portfolio choice; Knowledge of Economics and Finance, Financial Sophistication. JEL Classification: D91, G11, D80. ∗ Maarten C.J. van Rooij, Economics & Research Division, De Nederlandsche Bank, P.O. Box 98, 1000 AB, Amsterdam ([email protected]), Annamaria Lusardi, Department of Economics, Dartmouth College, Hanover, NH 03755 ([email protected]), and Rob J.M. Alessie, School of Economics and Business, University of Groningen, P.O. Box 800, 9700 AV, Groningen ([email protected]). We are grateful to James Banks, Johannes Binswanger, Marcello Bofondi, Henrik Cronqvist, Dimitris Georgarakos, Michael Haliassos, Lex Hoogduin, Adriaan Kalwij, Arie Kapteyn, Clemens Kool, Mauro Mastrogiacomo, Theo Nijman, Gerard van den Berg, Peter van Els, Arthur van Soest, and participants in the Program on the Global Demography of Aging Seminar Series at the Harvard School of Public Health, the CEPR/Netspar European Pension Challenges Conference (London, September 2006), the Netspar Workshop on the Micro-economics of Ageing (Utrecht, November 2006), the Plenary Session at the Italian Congress of Econometrics and Empirical Economics (Rimini, January 2007), the Workshop on Behavioral Approaches to Consumption, Credit, and Asset Allocation (European University Institute, May 2007), the Netspar Workshop (Groningen, June 2007), and the Conference on the Luxembourg Wealth Study: Enhancing Comparative Research on Household Finance (Rome, July 2007) for suggestions and comments. We also thank the staff of CentERdata and, in particular, Corrie Vis for their assistance in setting up the survey and the field work. The views expressed in this paper are those of the authors and do not necessarily reflect the views of De Nederlandsche Bank.
48
1. Introduction
Individuals have become increasingly active in financial markets, and market
participation has been accompanied or even promoted by the advent of new financial products
and services. However, some of these products are complex and difficult to grasp, especially
for financially unsophisticated investors. At the same time, market liberalization and
structural reforms in Social Security and pensions have caused an ongoing shift in decision
power away from the government and employers toward private individuals. Thus,
individuals have to assume more responsibility for their own financial well-being.
Are individuals well-equipped to make financial decisions? Do they possess adequate
financial literacy and knowledge? There has been little research on this topic and the few
existing studies indicate that financial illiteracy is widespread and individuals lack knowledge
of even the most basic economic principles (Lusardi and Mitchell, 2006, 2007a; National
Council on Economic Education (NCEE, 2005), and Hilgert, Hogarth and Beverly, 2003). At
the same time, there are concerns that households are not saving enough for retirement, are
accumulating excessive debt, and are not taking advantage of financial innovation (Lusardi
and Mitchell, 2007b; Campbell, 2006). The existing studies have also shown that those who
are not financially literate are less likely to plan for retirement and to accumulate wealth
(Lusardi and Mitchell, 2006, 2007a), and are more likely to take up high-interest mortgages
(Moore, 2003).
To measure financial literacy and assess its relationship with financial decision-
making, we have devised two special modules for the DNB Household Survey (DHS), a panel
data set covering a representative sample of the Dutch population and providing information
on savings and portfolio choice. We have designed an extensive list of questions aimed at
measuring and differentiating among different levels of literacy and financial sophistication.
These questions can be linked to a rich set of data on demographic characteristics and wealth
holdings. Our data show that the majority of households display basic financial knowledge
and have some grasp of concepts such as interest compounding, inflation, and the time value
of money. However, very few go beyond these basic concepts; many households do not know
the difference between bonds and stocks, the relationship between bond prices and interest
rates, and the basics of risk diversification. Most important, we find that financial literacy
affects financial decision-making: Those with low literacy are more likely to rely on family
and friends as their main source of financial advice and are less likely to invest in stocks.
This paper makes three contributions to the existing literature. First, we develop two
indices of financial literacy and knowledge, which allow us to differentiate among different
49
levels of financial sophistication. Adding this information to existing data sets can
substantially enhance the studies on saving and portfolio choice. Second, we contribute to the
methodology of measuring financial knowledge. There is a lot of noise in the responses to
financial literacy questions and we show that the wording of the questions is critically
important for measuring financial knowledge. Third, we provide a contribution toward
solving the so-called stock-holding puzzle, i.e., the fact that many households do not hold
stocks (Campbell, 2006; Haliassos and Bertaut, 1995). We show that many families shy away
from the stock market because they have little knowledge of stocks, the working of the stock
market, and asset pricing. To address the direction of causality between literacy and stock
market participation, we designed questions to measure not only current levels of literacy but
also levels of literacy in the past. Moreover, we designed questions to measure cognitive
ability in an attempt to disentangle the effects of knowledge from talents and skills.
Our findings have important policy implications. First, we show that financial literacy
should not be taken for granted. A majority of households possesses limited financial literacy.
Second, financial literacy differs substantially depending on education, age and gender. This
suggests that financial education programs are likely to be more effective when targeted to
specific groups of the population. Finally, any privatization programs should take into account
that, when put in charge of investing for their retirement, financially unsophisticated
individuals may not invest in the stock market. Thus, to work effectively, privatization
programs need to be accompanied by well-designed financial education programs.
This paper is organized as follows: In Section 2, we provide a review of the current
literature on financial literacy and stock market participation. In Section 3, we describe our
data set. In Section 4, we introduce our measures of financial literacy and describe the
problems of measuring literacy. In Section 5, we report the results of our empirical work. In
Section 6, we discuss our results and provide several extensions. In Section 7, we conclude
and examine areas for future research.
2. Literature review
There exist very few surveys that provide information on both financial literacy and
variables related to financial decision-making (for example saving, portfolio choice, and
retirement planning). To remedy this lack of data, Lusardi and Mitchell (2006) devised a
module on financial literacy for the 2004 US Health and Retirement Study (HRS). Their
questions aimed to test basic financial knowledge related to the working of interest
compounding, the effects of inflation, and risk diversification. They found that financial
50
illiteracy is widespread and particularly acute among specific groups of the population, such
as women, the elderly, and those with low education. These results are surprising not only
because the literacy questions were rather simple and basic, but also because their sample was
composed of respondents who are 50 or older. Most respondents in that age group have
checking accounts, credit cards, and have taken out one or two mortgages. However, similar
results are found in the work by Hilgert and Hogarth (2002), which examines financial
literacy in a sample covering all age groups, and on surveys by the National Council on
Economic Education (NCEE), that cover financial literacy among high school students and
the adult population. Findings of widespread illiteracy are also reported in studies on smaller
samples or specific groups of the population (Agnew and Szykman, 2005; Bernheim, 1995,
1998; Mandell, 2004; Moore, 2003).
While these studies focus on data from the US, surveys from other countries show
very similar results. A study by the OECD (2005) and work by Lusardi and Mitchell (2007b)
review the evidence on financial literacy across countries and show that financial illiteracy is
a common feature in many other developed countries, including European countries,
Australia, and Japan. These findings are echoed in the work of Christelis, Jappelli and Padula
(2007), which uses data very similar to the US HRS, and finds that most respondents in
Europe score low on numeracy scales.
Financial illiteracy has implications for household behavior. Bernheim (1995, 1998)
was the first to point out not only that most households cannot perform very simple
calculations and lack basic financial knowledge, but also that the savings behavior of many
households is dominated by crude rules of thumb. In more recent works, Bernheim, Garrett
and Maki (2001) and Bernheim and Garrett (2003) show that those who were exposed to
financial education in high school or in the workplace save more. Similarly, Lusardi and
Mitchell (2006, 2007a) show that those who display low literacy are less likely to plan for
retirement and as a result accumulate much less wealth (see also Hilgert, Hogarth and
Beverly, 2003). This finding is confirmed in the work by Stango and Zinman (2007), which
shows that those who are not able to correctly calculate interest rates out of a stream of
payments end up borrowing more and accumulating lower amounts of wealth. Agarwal,
Driscoll, Gabaix and Laibson (2007) further show that financial mistakes are prevalent among
the young and elderly, who are those displaying the lowest amount of financial knowledge.
The measures of financial literacy used in existing studies are often crude. For
example, Lusardi and Mitchell (2006, 2007a) rely on only three questions to measure
financial literacy, and Stango and Zinman (2007) rely on one question. Moreover, the surveys
51
that provide more extensive information about financial literacy often have little or no data on
wealth, saving, or other important economic outcomes (see, for example, the NCEE survey).
In this paper, we overcome the problems with some of the previous studies by providing
comprehensive measures of financial literacy as well as providing an evaluation of the quality
of the literacy data. In addition, we link financial literacy with an important economic
outcome: participation in the stock market. While extensive research on this topic exists, it is
still a puzzle why so many households do not hold stocks (Campbell, 2006). Some have
argued that short sale constraints, income risk, inertia, and departures from expected utility
maximization may explain why so few households hold stocks (Haliassos and Bertaut, 1995),
but it has proven hard to account for all these factors in available micro data sets. Others have
argued that young people cannot borrow and thus do not have wealth to invest in stocks
(Constantinides, Donaldson and Mehra, 2002). These life-cycle considerations and the wedge
between borrowing and lending rates can provide some explanation for lack of stock
ownership (Davis, Kubler and Willen, 2006), but even these reasons cannot fully explain why
such a large proportion of families do not hold stocks. More recent papers have incorporated
other reasons, such as trust and culture (Guiso, Sapienza and Zingales, 2005), and the
influence of neighbors and peers (Hong, Kubik and Stein, 2004; Brown, Ivkovic, Smith and
Weisbenner, 2007). Yet other authors have started to consider limited numeracy and cognitive
ability (Christelis, Jappelli and Padula, 2007), lack of asset awareness (Guiso and Jappelli,
2005), and lack of financial sophistication (Kimball and Shumway, 2006). Our work
improves substantially upon these studies by considering more refined indices of financial
literacy and financial sophistication that we have explicitly designed for a survey of Dutch
households. Moreover, to better understand the relationship between financial literacy and
stock market participation, we have designed questions to measure economic knowledge
before entering the stock market.
3. Data
We use data from the 2005 DNB Household Survey (DHS). DHS is an annual
household survey covering information about demographic and economic characteristics and
focusing on wealth and saving data. The panel is run by CentERdata, a survey research
institute at Tilburg University that specializes in internet surveys.1 The data set is
representative of the Dutch population, and it contains over 2000 households.
1 http://www.uvt.nl/centerdata/en/. See Nyhus (1996) for a detailed description of this survey and an assessment of the quality of the data.
52
In addition to using data from the main core of the DHS, we also use data from two
modules we designed, which were added to the survey in 2005 and 2006. The first financial
literacy module was in the field from September 23 until September 27, 2005 and was
repeated a week later for those who did not respond during that time. A total of 1508 out of
2028 households completed the financial literacy module, implying a response rate of 74.4%
(in line with the response rate from the main survey). A second module was fielded in January
2006, and 1373 out of the original 1508 respondents completed that module. The respondent
to the financial literacy questions is the member of the household in charge of household
finances.
Survey participants are interviewed via the internet. Although the internet connection
rate in the Netherlands is one of the highest in Europe (80% of Dutch households are
connected to the internet at their home), households need not have an internet connection to
participate in the survey. Recruitment and selection of households is first done by phone with
a randomly selected sample of households. Households without an internet connection are
provided with a connection or with a set-top box for their television (for those who do not
have access to a personal computer). This method of data collection presents several
advantages. For example, data collected with internet surveys suffer less from reporting biases
than those collected via telephone interviews (Chang and Krosnick, 2003).
The age of the respondents in our sample varies from 22 to 90 (mean age is 49.6);
51.5% of respondents are male; 34.5% have a college education (which includes vocational
training in addition to university degrees). In regards to household composition, 56.8% of
respondents are married or living together with a partner, and one third have children living at
home. Overall, 18.4% of respondents are retired (including early retirees), 10.8 % are disabled
or unemployed, and 4.4% are self-employed.2
4. The measurement of literacy
As mentioned before, we designed two modules to measure and evaluate financial
literacy. The financial literacy questions are composed of two parts. The first set of questions
aims to assess basic financial literacy. These questions cover topics ranging from the working
of interest rates and interest compounding to the effect of inflation, discounting and nominal
versus real values. The second set of questions aims to measure more advanced financial
knowledge and covers topics such as the difference between stocks and bonds, the function of
2 Throughout our empirical analysis, we always use household weights to ensure that our statistics are representative of the population.
53
the stock market, the working of risk diversification, and the relationship between bond prices
and interest rates. These questions were designed using similar modules in the HRS and a
variety of other surveys on financial literacy. However, a few questions are unique to our
module on literacy.3 Households are instructed to answer the questions without consulting
additional information or using a calculator.4
The exact wording of the questions measuring basic financial literacy is reported
below in Box 1:
Box 1. Basic Literacy Questions 1) Numeracy Suppose you had €100 in a savings account and the interest rate was 2% per year. After 5 years, how much do you think you would have in the account if you left the money to grow? (i) More than €102; (ii) Exactly €102; (iii) Less than €102; (iv) Do not know; (v) Refusal. 2) Interest compounding Suppose you had €100 in a savings account and the interest rate is 20% per year and you never withdraw money or interest payments. After 5 years, how much would you have on this account in total? (i) More than €200; (ii) Exactly €200; (iii) Less than €200; (iv) Do not know; (v) Refusal. 3) Inflation Imagine that the interest rate on your savings account was 1% per year and inflation was 2% per year. After 1 year, how much would you be able to buy with the money in this account? (i) More than today; (ii) Exactly the same; (iii) Less than today; (iv) Do not know; (v) Refusal. 4) Time value of money Assume a friend inherits €10000 today and his sibling inherits €10000 3 years from now. Who is richer because of the inheritance? (i) My friend; (ii) His sibling; (iii) They are equally rich; (iv) Do not know; (v) Refusal. 5) Money illusion Suppose that in the year 2010, your income has doubled and prices of all goods have doubled too. In 2010, how much will you be able to buy with your income? (i) More than today; (ii) The same; (iii) Less than today; (iv) Do not know; (v) Refusal.
These questions measure the ability to perform simple calculations (in the first
question), the understanding of how compound interest works (second question), and the
effect of inflation (third question). We also designed questions to assess the knowledge of
3 For an analysis of the module on financial literacy in the 2004 HRS, see Lusardi and Mitchell (2006). For a review of financial literacy surveys across countries, see Lusardi and Mitchell (2007b). 4 This facilitates the comparison with other surveys, which are normally done via telephone. Moreover, this procedure better enables researchers to assess what respondents know.
54
time discounting (fourth question) and whether respondents suffer from money illusion (fifth
question). These concepts lie at the basis of basic financial transactions, financial planning,
and day-to-day financial decision-making.
Responses to these questions are reported in Table 1A. Most respondents answer the
first question correctly, where the percentage of incorrect responses is only 5.2%. However,
the proportion of correct answers decreases considerably, to a little more than 70%, when we
consider questions on interest compounding, time discounting, and money illusion; the
proportion of incorrect answers on questions measuring the time value of money or money
illusion is around 24%. Note also that, while many respondents answer each individual
question correctly, the proportion of respondents who answered all five questions correctly is
only 40.2% (Table 1B). Thus, while many respondents display knowledge of a few financial
concepts, basic financial literacy is not widespread.
Table 1A. Basic financial literacy Weighted percentages of total number of respondents (N=1508)
____________________________________________________________________________ Numeracy Interest
compounding Inflation Time value
of money Money illusion
___________ ___________ ___________ ___________ ___________ Correct 90.8 76.2 82.6 72.3 71.8 Incorrect 5.2 19.6 8.6 23.0 24.3 Do not know 3.7 3.8 8.5 4.3 3.5 ____________________________________________________________________________ Note: Correct, incorrect, and do not know responses do not sum up to 100% because of refusals.
Table 1B. Basic literacy: Summary of responses Weighted percentages of total number of respondents (N=1508)
____________________________________________________________________________ Number of correct, incorrect and do not know answers (out of five
questions) _____________________________________________________________ None 1 2 3 4 All Mean _____ _____ _____ _____ _____ _____ _____ Correct 2.3 2.8 6.7 15.1 32.8 40.2 3.94 Incorrect 45.2 35.7 13.6 4.4 1.1 0.0 0.81 Do not know 88.9 5.9 1.7 1.4 0.7 1.5 0.24 ____________________________________________________________________________ Note: Categories do not sum up to 100% because of rounding and means do not sum up to 5 due to refusals.
55
To be able to classify respondents according to different levels of financial
sophistication, we added several other questions to the module. The exact wording of these
questions is reported in Box 2.
Box 2. Advanced Literacy Questions 6) Which of the following statements describes the main function of the stock market? (i) The stock market helps to predict stock earnings; (ii) The stock market results in an increase in the price of stocks; (iii)The stock market brings people who want to buy stocks together with those who want to sell stocks; (iv) None of the above; (v) Do not know; (vi) Refusal. 7) Which of the following statements is correct? If somebody buys the stock of firm B in the stock market: (i) He owns a part of firm B; (ii) He has lent money to firm B; (iii) He is liable for firm B’s debts; (iv) None of the above; (v) Do not know; (vi) Refusal. 8) Which of the following statements is correct? (i) Once one invests in a mutual fund, one cannot withdraw the money in the first year; (ii) Mutual funds can invest in several assets, for example invest in both stocks and bonds; (iii) Mutual funds pay a guaranteed rate of return which depends on their past performance; (iv) None of the above; (v) Do not know; (vi) Refusal. 9) Which of the following statements is correct? If somebody buys a bond of firm B: (i) He owns a part of firm B; (ii) He has lent money to firm B; (iii) He is liable for firm B’s debts; (iv) None of the above; (v) Do not know; (vi) Refusal. 10) Considering a long time period (for example 10 or 20 years), which asset normally gives the highest return? (i) Savings accounts; (ii) Bonds; (iii) Stocks; (iv) Do not know; (vi) Refusal. 11) Normally, which asset displays the highest fluctuations over time? (i) Savings accounts; (ii) Bonds; (iii) Stocks; (iv) Do not know; (v) Refusal. 12) When an investor spreads his money among different assets, does the risk of losing money: (i) Increase; (ii) Decrease; (iii) Stay the same; (iv) Do not know; (v) Refusal. 13) If you buy a 10-year bond, it means you cannot sell it after 5 years without incurring a major penalty. True or false? (i) True; (ii) False); (iii) Do not know; (iv) Refusal. (14) Stocks are normally riskier than bonds. True or false? (i) True; (ii) False; (iii) Do not know; (iv) Refusal. (15) Buying a company stock usually provides a safer return than a stock mutual fund. True or false? (i) True; (ii) False; (iii) Do not know; (iv) Refusal. (16) If the interest rate falls, what should happen to bond prices? (i) Rise; (ii) Fall; (iii) Stay the same; (iv) None of the above; (v) Do not know; (vi) Refusal.
56
Clearly, these are much more complex questions than the previous set.5 The purpose
of these questions is to measure more advanced financial knowledge related to investment and
portfolio choice. Specifically, these questions were devised to assess knowledge of financial
assets, such as stocks, bonds and mutual funds, the returns and riskiness of different assets, as
well as the working of the stock market. Moreover, we attempt to measure whether
respondents understand the concept of risk diversification (which was asked in two separate
questions), the working of mutual funds, and the relationship between bond prices and interest
rates.
Reponses to these questions are reported in Table 2A. The pattern of answers is much
different than in the previous set of questions. For example, the proportion of correct answers
on each question is much lower; only a quarter of respondents know about bond pricing and
only 30% know how long-term bonds work. Respondents also display difficulties in grasping
the concept of risk diversification: Less than 50% of respondents know that a stock mutual
fund is safer than a company stock. Not only do a sizable proportion of respondents answer
these questions incorrectly, but also many respondents state they do not know the answers to
these questions. For example, while 30% of respondents are incorrect about which asset
(among savings accounts, bonds and stocks) gives the highest return over a long time period,
an additional 22% do not know the answer to this question. Similarly, more than 37% are
incorrect about the relationship between bond prices and interest rates and the same high
percentage (37.5%) state they do not know the answer to that question. Many respondents are
incorrect or do not know the definition of stocks, bonds, and the working of mutual funds.
Table 2B shows that only a tiny fraction of respondents (5%) are able to answer all the
advanced literacy questions correctly, while the fraction of incorrect responses or ‘do not
know’ answers on several questions is sizable. These are important findings; most models of
portfolio choice assume that investors are knowledgeable and well-informed. Instead, the
findings in Tables 1A, 1B, 2A, and 2B show that financial literacy should not be taken for
granted. These findings echo the results found in US surveys, such as the HRS and the Survey
of Consumers (Lusardi and Mitchell, 2006; Hilgert, Hogarth and Beverly, 2003). 5 Because we could not perform a pilot study to assess how respondents perform on these questions and how well they understood them, we use the wording of questions from other existing surveys (with some modifications to reflect the characteristics of the Dutch financial system and the behaviour of Dutch financial markets). Specifically, we took question 6 from the National Council of Economic Education Survey, questions 7 and 9 from the NASD Investor Knowledge Quiz, question 15 from the 2004 Health and Retirement Study module on financial literacy, questions 8, 10, 11, 12, 13, 14 and 16 from the Survey of Financial Literacy in Washington State, the Survey of Consumers, and the John Hancock Financial Services Defined Contribution Plan Survey. We took the questions that best reflect financial sophistication related to financial instruments and the working of the stock market. As explained later, we have also experimented with the wording of some of these questions.
57
Table 2A. Advanced financial literacy Weighted percentages of total number of respondents (N=1508)
___________________________________________________________________________________ Correct Incorrect DK ______ _______ ____ Which statement describes the main function of the stock market? 1)
67.0 12.9 19.7
What happens if somebody buys the stock of firm B in the stock market?1)
62.2 25.7 11.0
Which statement about mutual funds is correct? 1)
66.7 11.1 21.7
What happens if somebody buys a bond of firm B? 1)
55.6 17.8 26.4
Considering a long time period (for example 10 or 20 years), which asset normally gives the highest return: savings accounts, bonds or stocks?
47.2 30.1 22.3
Normally, which asset displays the highest fluctuations over time: savings accounts, bonds, stocks?
68.5 12.7 18.4
When an investor spreads his money among different assets, does the risk of losing money increase, decrease or stay the same?
63.3 17.4 19.0
If you buy a 10-year bond, it means you cannot sell it after 5 years without incurring a major penalty. True or false?
30.0 28.3 37.9
Stocks are normally riskier than bonds. True or false? 2)
60.2 15.1 24.3
Buying a company fund usually provides a safer return than a stock mutual fund. True or false? 2)
48.2 24.8 26.6
If the interest rate falls, what should happen to bond prices: rise/fall/stay the same/none of the above? 2)
24.6 37.1 37.5
___________________________________________________________________________________ 1) See exact wording in Box 2. 2) This question has been phrased in two different ways. See also Table 3. Note: DK = ‘Do not know’; Correct, incorrect and DK responses do not sum up to 100% because of refusals.
Table 2B. Advanced literacy: Summary of responses Weighted percentages of total number of respondents (N=1508)
__________________________________________________________________________________ Number of correct, incorrect and do not know answers (out of eleven questions) _________________________________________________________________________ None 1 2 3 4 5 6 7 8 9 10 All Mean ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ ____ Correct 7.6 5.1 5.2 6.4 7.3 10.0 11.1 11.3 10.8 10.6 9.8 5.0 5.93 Incorrect 18.7 20.2 19.8 16.8 10.4 7.1 4.7 1.6 0.6 0.1 0.0 0.0 2.33 DK 44.2 11.4 8.0 6.1 5.1 3.7 4.1 4.2 2.8 3.2 3.5 3.6 2.65 __________________________________________________________________________________ Note: DK = ‘Do not know’; Categories do not sum up to 100% because of rounding and means do not sum up to 11 due to refusals.
58
When lack of financial knowledge is so widespread, one has to worry about whether
respondents even understood the meaning of the questions, and the prevalence of guessing
and random answers. To assess the relevance of these problems, we used the following
strategy: We inverted the wording of questions and exposed two randomly chosen groups of
respondents to the same question but with a different wording. We did so for three types of
questions: A simple question about the riskiness of bonds versus stocks, a more difficult
question about the riskiness of a company stock versus a stock mutual fund, and an even more
complex question on the effect of interest rate changes on bond prices. This allows us to
assess how incorrect and perhaps random answers are connected to the difficulty of the
questions. The precise wording of the questions is reported below:
(14a) Stocks are normally riskier than bonds. True or false? (14b) Bonds are normally riskier than stocks. True or false? (15a) Buying a company stock usually provides a safer return than a stock mutual fund. True or false? (15b) Buying a stock mutual fund usually provides a safer return than a company stock. True or false? (16a) If the interest rate falls, what should happen to bond prices? Rise/fall/stay the same/none of the above? (16b) If the interest rate rises, what should happen to bond prices? Rise/fall/stay the same/none of the above?
The pattern of responses in Table 3 shows that the wording of the question matters,
particularly for the difficult questions. When comparing the response to a simple question on
the riskiness of stocks versus bonds, we find that respondents give rather similar answers
regardless of the wording of the question (differences are not significant at the 5% level of
significance). However, this is not the case for complex questions. The pattern of answers
changes dramatically when the order of the wording was inverted. For example, the number
of correct answers doubles when respondents are asked whether ‘buying a company stock
usually provides a safer return than a stock mutual fund’ versus the same question with the
inverted order: ‘buying a stock mutual fund provides a safer return than a company stock’.
Note that this is not the result of following a crude rule of thumb, such as picking the first
answer as the correct one. This would lead to a lower rather than higher percentage of correct
answers for question (15a).6 This finding provides evidence that respondents often do not
6 It is consistent, however, with another rule of thumb that was mentioned to us about the behavior of students. They tend to reply ‘false’ to a true-false question when they are not sure about the answer.
59
understand the question or do not know what stocks, bonds, and mutual funds are, and some
correct answers are simply the result of guessing. It also shows that answers to advanced
financial literacy questions should not be taken at face value and the empirical work should
take into account that these measures are often noisy proxies of the true level of financial
knowledge. We will address these issues in the empirical work.
Table 3. Advanced literacy: Responses to questions with inverted wording Weighted percentages
__________________________________________________________________________________ Correct Incorrect DK ______ _______ ______ Stocks are normally riskier than bonds. True or false? (N=751) 60.8 17.1 21.7 Bonds are normally riskier than stocks. True or false? (N=757) 59.7 13.1 26.9 Pearson chi2(2) = 5.25 (p = 0.072) __________________________________________________________________________________ Buying a company stock usually provides a safer return than a stock mutual fund. True or false? (N=763)
63.4 12.1 24.1
Buying a stock mutual fund usually provides a safer return than a company stock. True or false? (N=745)
32.3 38.1 29.2
Pearson chi2(2) = 184.59 (p = 0.000) __________________________________________________________________________________ If the interest rate falls, what should happen to bond prices: rise/fall/stay the same/none of the above? (N=755)
30.5 33.8 34.8
If the interest rate rises, what should happen to bond prices: rise/fall/stay the same/none of the above? (N=753)
18.9 40.3 40.3
Pearson chi2(2) = 23.15 (p = 0.000) __________________________________________________________________________________ Note: DK = ‘Do not know’; Correct, incorrect, and do not know responses do not sum up to 100% because of refusals. In performing the test, we group together ‘do not know’ and ‘refusal’ responses.
4.1 Indices of financial literacy
We summarize all of the information about financial literacy resulting from our two
sets of questions into a financial literacy index. We first combine the information we have
available by performing a factor analysis on the sixteen questions in the financial literacy
module. Consistent with the way we have devised the financial literacy questions, the factor
analysis indicates there are two main factors with different loading on two types of questions:
The simple literacy questions (first 5 questions) and the more advanced literacy questions
(remaining 11 questions). We decided therefore to split the set of questions into two groups
and perform a factor analysis on the two sets separately. In this way, we can construct two
types of literacy indices: a first literacy index potentially related to basic knowledge (note that
there are no questions in this set about the stock market or about stocks and bonds) and a
second index measuring more advanced financial knowledge as well as knowledge related to
60
stocks, the stock market and other financial instruments. In constructing the indices, we
explicitly take into account the differences between ‘incorrect’ answers and ‘do not know’
answers. As already reported in Lusardi and Mitchell (2006), it is important to exploit this
information to differentiate among degrees of financial knowledge. Details about the factor
analysis are reported in Appendix A.
The basic literacy index runs from a minimum value of -2.9 for respondents without
any correct answer to a maximum of 1.0 for the participants with only correct responses. The
advanced literacy index goes from -4.7 to 0.8. Both distributions have mean zero and a
standard deviation of 1.0 and 1.2 respectively. As expected the basic and advanced literacy
measures are clearly correlated albeit far from perfect (correlation coefficient: 0.46).
To confirm the validity of these two indices and their features, we report the
distribution of the financial literacy indices across demographic variables such as education,
age, and gender in Tables 4A and 4B. As expected, basic financial literacy increases strongly
with education. Those with the lowest level of basic financial literacy are concentrated on the
lowest education categories: primary and preparatory intermediate vocational schools.
Conversely, those with a higher vocational education (similar to a college degree in the US)
or a university education locate in the highest quartiles of the basic literacy index. The profile
of basic literacy has a hump-shape with regards to age, although not very pronounced. Even
though in a single cross-section we cannot distinguish between age and cohort effects, this
finding is similar to what is reported in Agarwal, Driscoll, Gabaix and Laibson (2007). Table
4A also shows there are large differences in basic literacy between gender: Women display
much lower basic knowledge than men. These findings are similar to those reported by
Lusardi and Mitchell (2006) and the findings in other literacy surveys (Lusardi and Mitchell,
2007b).
Considering more advanced financial knowledge in Table 4B, again we find a strong
relationship with education. A large fraction (48.3%) of respondents with primary education
is at the lowest level of literacy (first quartile). As we move to higher quartiles of level of
literacy, the proportion of respondents with high levels of education increases, but even when
we consider those with a university degree, only 43.4%% of them are at the top quartile of
advanced literacy (the proportion was 70.9% when we consider basic literacy). Thus, even
respondents with high educational attainment can display a low degree of financial knowledge
(more than 30% of respondents with a university degree are in the bottom two quartiles of the
advanced literacy index distribution). Thus, while strongly correlated, education is only an
61
Table 4A. Basic literacy across demographics Weighted percentages
_______________________________________________________________________________ Basic literacy quartiles _____________________________________________ Education 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ Primary 35.8 31.1 17.1 15.9 2.13 67 Preparatory intermediate voc. 30.5 22.7 21.8 25.0 2.41 345 Intermediate vocational 20.9 20.8 25.2 33.2 2.71 294 Secondary pre-university 11.1 20.8 25.7 42.4 2.99 207 Higher vocational 6.4 18.1 24.0 51.5 3.21 397 University 5.9 9.7 13.5 70.9 3.49 197 Pearson chi2(15) = 147.42 (p=0.000) _______________________________________________________________________________ Basic literacy quartiles _____________________________________________ Age 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ 21-30 years 21.6 19.7 19.4 39.4 2.76 179 31-40 years 18.8 18.3 21.1 41.9 2.86 306 41-50 years 13.7 18.0 23.9 44.3 2.99 333 51-60 years 16.6 19.8 21.3 42.3 2.89 311 61-70 years 18.3 22.3 23.8 35.6 2.77 217 71 years and older 18.3 24.1 24.6 33.0 2.72 162 Pearson chi2(15) = 12.23 (p=0.661) _______________________________________________________________________________ Basic literacy quartiles _____________________________________________ Gender 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ Female 22.2 25.4 21.2 31.2 2.62 674 Male 13.3 14.9 23.2 48.6 3.07 834 Pearson chi2(3) = 52.99 (p=0.000) _______________________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
imperfect proxy for financial literacy and empirical studies that account for education may not
fully account for the effect of financial knowledge.
Advanced literacy is low among the young, is highest among middle-age respondents
(particularly 40 to 60), and declines slightly at an advanced age (61 or older). This suggests
that people may be learning as they age and, perhaps, participate in financial markets. Gender
differences become even sharper when considering advanced literacy. A large percentage of
62
Table 4B. Advanced literacy across demographics Weighted percentages
_______________________________________________________________________________ Advanced literacy quartiles _____________________________________________ Education 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ Primary 48.3 24.7 17.5 9.5 1.88 67 Preparatory intermediate voc. 35.1 29.4 23.5 12.0 2.12 345 Intermediate vocational 32.8 23.9 26.3 17.0 2.28 294 Secondary pre-university 19.0 21.8 28.4 30.9 2.71 207 Higher vocational 14.6 23.7 25.1 36.7 2.84 397 University 6.0 24.7 26.0 43.4 3.07 197 Pearson chi2(15) = 149.32 (p=0.000) _______________________________________________________________________________ Advanced literacy quartiles _____________________________________________ Age 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ 21-30 years 24.0 33.5 25.4 17.1 2.36 179 31-40 years 34.3 21.3 23.5 20.9 2.31 306 41-50 years 23.4 26.5 20.5 29.7 2.56 333 51-60 years 18.2 24.1 30.6 27.1 2.67 311 61-70 years 25.7 22.5 22.2 29.6 2.56 217 71 years and older 23.2 24.1 28.7 24.1 2.54 162 Pearson chi2(15) = 36.70 (p=0.001) _______________________________________________________________________________ Advanced literacy quartiles _____________________________________________ Gender 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ Male 15.9 20.2 26.7 37.2 2.85 834 Female 34.5 30.2 23.3 12.1 2.13 674 Pearson chi2(3) = 161.53 (p=0.000) _______________________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
women display low literacy: 34.5% of women are in the first and lowest quartile of the
literacy distribution while only 12.1% are at the fourth quartile; the corresponding figures for
men are 15.9% and 37.2% respectively.
To further show that these indices measure economic knowledge, in Table 4C we
report the relationship between these measures of literacy and a subjective measure of
financial knowledge. In our module we have asked respondents to report on a scale from 1 to
63
Table 4C. Basic and advanced literacy versus self-assessed literacy Weighted percentages
_______________________________________________________________________________ Basic literacy quartiles _____________________________________________ Self-assessed literacy 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ 1 (very low) 29.6 30.4 16.2 23.8 2.34 9 2 15.1 26.4 13.0 45.5 2.89 56 3 28.6 19.9 24.8 26.7 2.50 137 4 20.4 23.6 18.7 37.4 2.73 366 5 15.5 19.7 25.3 39.6 2.89 499 6 8.6 16.9 22.2 52.3 3.18 355 7 (very high) 7.4 13.4 25.5 53.7 3.25 45 Do not know 53.4 12.7 18.5 15.5 1.96 31 Refusal 52.9 0.0 35.9 11.2 2.05 10 Pearson chi2(24) = 100.38 (p=0.000) _______________________________________________________________________________ Advanced literacy quartiles _____________________________________________ Self-assessed literacy 1 (low) 2 3 4 (high) Mean N ________________________ ________ ________ ________ ________ ________ ________ 1 (very low) 55.3 9.4 27.1 8.2 1.88 9 2 24.9 34.9 22.2 18.0 2.33 56 3 29.2 31.8 28.1 10.9 2.21 137 4 31.3 27.5 23.2 18.0 2.28 366 5 21.7 28.1 25.8 24.4 2.53 499 6 15.9 15.6 26.1 42.4 2.95 355 7 (very high) 3.9 10.2 34.8 51.1 3.33 45 Do not know 66.1 18.3 8.6 7.0 1.56 31 Refusal 67.5 24.9 7.6 0.0 1.40 10 Pearson chi2(24) = 189.19 (p=0.000) _______________________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
7 their understanding of economics.7 Such a question has the advantage of being simple and
direct. Moreover, it does not mention stock market participation. Note also that the question
was located at the beginning of the literacy module, before any of the questions included in
the basic and advanced financial literacy indices were asked. Thus, respondents had to assess
their own knowledge before they answered the literacy questions. Most respondents assessed
their economic knowledge as being above 3: 25.38% of respondents stated their level is 4,
32.75% that their level is 5 and 24.27% that their level is 6. However, only 2.71% reported
7 See appendix B for the precise wording of this question.
64
their knowledge of economics as being very high (7). Most importantly, there is a very strong
correlation between objective and subjective literacy. More than 50% of respondents who
report knowing a lot about economics (score of 6 or 7) are located in the top quartile of the
basic literacy index. The relationship becomes even stronger when we consider the advanced
literacy index. More than 50% of respondents who report low levels of economic knowledge
(score of 1, 2 or 3) are located in the first two quartiles of the literacy index, while the
majority of those with high knowledge are located in the top two quartiles of the literacy
index. Thus, while there may be noise and measurement error affecting these indices, they do
provide information about economic knowledge.
An important question we aim to answer in our paper is not only whether respondents
possess financial literacy, but also whether financial literacy matters in financial decision-
making. We do so by first examining whether literacy influences the sources of information
households consult when making financial decisions, to shed some light on why literacy
affects financial behavior. We then examine whether financial literacy affects participation in
the stock market.
Table 5 shows that a high proportion of respondents with low basic literacy rely on
informal sources of information, such as family, friends and acquaintances. However, this
proportion sharply decreases when we move to higher levels of basic literacy. Conversely, the
proportion of households relying on newspapers, financial magazines, guides and books, and
financial information on the Internet increases substantially as we move from low levels of
literacy to high levels of basic literacy. Households with higher financial literacy are also
more likely to rely on professional financial advisers. The effect is similar but stronger when
we look at advanced financial literacy. Those who display high levels of advanced literacy are
much less likely to rely on informal sources of information such as family and friends, and
much more likely to read newspapers and magazines, consult financial advisors, and seek
information on the Internet. While correlation does not imply causation, this table shows that
financial literacy is strongly connected with sources of financial advice. Insofar as financial
advice is an input in financial-decision making and leads to better saving and investment
decisions, the findings reported in Table 5 provide a reason why financial literacy matters. In
the next section, we look directly at financial behavior by examining whether financial
literacy has an effect on stock market participation.
65
Table 5. Most important source of advice for different levels of literacy Weighted percentages (N=1135)
_________________________________________________________________________________ Basic literacy quartiles _______________________________________
What is your most important source of advice when you have to make important financial decisions for the household? 1 (low) 2 3 4 (high) _______________________________________ _________ _________ _________ _________ - Parents, friends or acquaintances 40.2 34.4 28.8 20.8 - Information from the newspapers 3.6 7.8 8.9 9.5 - Financial magazines, guides, books 3.9 7.5 9.3 12.4 - Brochures from my bank or mortgage adviser 10.6 6.8 6.0 8.1 - Advertisements on TV, in papers or other media 3.7 3.2 2.8 3.9 - Professional financial advisers 21.8 21.3 24.2 25.5 - Financial computer programs 0.0 0.3 0.9 0.7 - Financial information on the Internet 4.0 7.5 8.1 10.5 - Other 12.3 11.4 11.0 8.6 _________________________________________________________________________________
Advanced literacy quartiles _______________________________________
What is your most important source of advice when you have to make important financial decisions for the household? 1 (low) 2 3 4 (high) _______________________________________ _________ _________ _________ _________ - Parents, friends or acquaintances 40.7 37.4 19.9 17.9 - Information from the newspapers 1.1 6.0 10.6 13.7 - Financial magazines, guides, books 2.1 7.6 9.7 17.0 - Brochures from my bank or mortgage adviser 6.6 6.7 11.3 6.2 - Advertisements on TV, in papers or other media 4.0 3.6 5.0 1.4 - Professional financial advisers 19.4 23.6 27.5 24.1 - Financial computer programs 0.2 0.3 1.1 0.5 - Financial information on the Internet 6.3 6.6 7.6 12.4 - Other 19.7 8.2 7.3 6.9 _________________________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
5. Financial literacy and stock market participation
As mentioned before, an important puzzle in the literature is why so few households
hold stocks. In our sample, 23.8% of households own stocks or mutual funds. Thus, as in the
US, many households do not participate in the stock market. This figure, however, hides
major differences among demographics groups. As reported in Table 6, stock ownership
increases sharply with education levels.8 Only a small fraction of those with low education
own stocks. However, even the large majority of those with a university degree do not
participate in the stock market. Thus, impediments to stock ownership go beyond levels of
schooling. Note that we found similar results when considering the index of basic and
8 Note that by merging the data on stock market participation and the financial literacy module, our sample reduces to 1,189 observations. However, we do not find evidence that our sample suffers from selectivity.
66
Table 6. Stock market participation across subgroups Weighted percentages (N=1189)
__________________________________________________________________________________ Education Age Primary 11.3 21-30 years 14.4 Preparatory intermediate voc. 16.0 31-40 years 19.4 Intermediate vocational 19.1 41-50 years 27.1 Secondary pre-university 22.5 51-60 years 26.8 Higher vocational 33.7 61-70 years 24.3 University 38.8 71 years and older 30.1 Gender Marital status Female 16.7 Not-married 19.8 Male 30.3 Married 26.8 Net household income quartiles Non-equity net wealth quartiles 1 (low) 13.4 1 (low) 7.1 2 17.5 2 20.3 3 29.1 3 29.7 4 (high) 35.9 4 (high) 37.9 Basic literacy quartiles Advanced literacy quartiles 1 (low) 7.7 1 (low) 7.5 2 21.2 2 15.0 3 22.0 3 26.5 4 (high) 32.8 4 (high) 44.4 __________________________________________________________________________________ Note: Stock market participation is defined as owning individual stocks and/or mutual funds.
advanced literacy; even those with high levels of schooling did not always score high on
financial knowledge. This suggests that schooling is not necessarily a good proxy for literacy
and models of portfolio choice may need to incorporate both variables to explain behavior
toward stocks. Stock market participation increases with age/cohorts; stock ownership is
concentrated among those 40 and older. The large proportion of stock ownership for those
older than 70 may simply be the result of differential mortality between richer and poorer
households (Hurd, 1990). Stock market participation is much lower among women than men,
a finding also reported in other studies (see also Haliassos and Bertaut, 1995) and consistent
with the sharp differences in literacy between women and men (Lusardi and Mitchell, 2006).
Stock market participation increases strongly with both income and wealth levels. Income
refers to household net disposable income: It is simply household total income (which is the
sum of labor income, unemployment and disability payments, social security an pension,
other transfers and capital income, minus taxes). Wealth is the sum of checking and savings
accounts, employer-sponsored savings plans, cash value of life insurance, home equity, other
67
real estate and other financial assets, minus total debt.9 These findings are similar to those
reported in many other papers on stock-ownership (see the review in Guiso, Haliassos and
Jappelli (2002) and Campbell (2006)).
One explanation about lack of stock ownership that has not yet been well-explored in
the literature is that stocks are complex assets, and many households may not know or
understand stocks and the working of the stock market. At the bottom of Table 6, we report
stock ownership across different levels of financial literacy. Stock ownership increases
sharply with literacy. Even when considering basic literacy that measures simple knowledge
and ability to do calculations, we find that those who score high on basic literacy are
disproportionately more likely to participate in the stock market. The relationship becomes
much stronger when we consider the index of advanced literacy. Participation in the stock
market is concentrated among those with high literacy (fourth quartile), while only 8% and
15% of respondents in the first and second quartile of literacy participate in the stock market.
Given that literacy is highly correlated with the demographic variables mentioned above, we
now turn to examine whether this relationship holds true even after accounting for many of
the determinants of stock market participation, such as age, education, gender, income and
wealth. Most important, we will address the direction of causality between stock ownership
and financial literacy.
Our empirical specification recognizes there are many determinants of stock
ownership, and we consider a wide set of variables that are available in our survey. As in the
previous studies, we consider demographics such as age, education, gender, marital status,
and number of children (Haliassos and Bertaut, 1995; Guiso, Haliassos and Jappelli, 2002;
Campbell, 2006). We added a dummy for respondents who are retired to account for the fact
that some households may be in the decumulation phase of their life-cycle. We also added a
dummy for self-employment, to account for those who are already exposed to high risk in the
labor market and may therefore be less likely to hold stocks (Heaton and Lucas, 2000).
Additionally, we added income (in logs) and dummies for quartiles of wealth.10 Most
important, we added measures of financial literacy. One of the main hypotheses of this paper
is that respondents who are not financially knowledgeable—do not know about stocks and
9 Because the dependent variable in our empirical work is stock market participation (including participation in mutual funds), in our definition of wealth we do not include stocks and mutual funds (which are clearly correlated with stock market participation). We also do not include business equity because it is a very noisy measure of business wealth. For an analysis of wealth and wealth components in the DHS, see Alessie, Hochguertel and van Soest (2002). 10 Wealth measures are rather noisy in the DHS. The use of dummies allows us to overcome this problem and also to measure how much stock-ownership increases over the wealth distribution.
68
bonds and are not familiar with the working of financial markets—stay away from the stock
market. We use the index for advanced literacy to account for financial knowledge. However,
we also add the index of basic knowledge to account for different levels of literacy as well as
to control for cognitive ability.11
Table 7 reports the results for several stock market participation regressions using both
OLS and GMM estimation techniques. We start by discussing the OLS results. As a
benchmark, the first column shows the empirical estimates for a traditional specification
without including direct measures of literacy. The estimates are in line with the results that
commonly found in the literature (Guiso, Haliassos and Jappelli, 2002). In particular,
education, gender, income and wealth explain the variation in stock ownership.
The second and third column of Table 7 show that financial literacy matters for stock
ownership, even after controlling for a large set of demographic characteristics and income
and wealth. Those who display higher literacy are more likely to participate in the stock
market. The estimates are also sizable: A one-standard deviation increase in advanced literacy
raises stock market participation by more than 8 percentage points. Note that the effect is as
large as the effect of formal education and wealth. For example, having a university degree
increases stock market participation by more than 9 percentage points. Compared to the first
quartile of wealth (values up to €2300), having wealth in the second quartile (up to €45000)
increases stock market participation by more than 7 percentage points. Note also that when
we account for basic literacy the estimate of advanced literacy does not change (column (3) of
Table 7). These estimates indicate that financial literacy affects stock market participation
above and beyond the effect of the traditional determinants of stock ownership. Compared to
the traditional specification, a larger part of the variation in stock ownership can be explained
and in particular the importance of education for stock ownership is reduced considerably (the
education dummies even become jointly insignificant) which suggests that this variable serves
as a proxy for financial literacy when excluding direct measures for financial ability and
knowledge.
There are several potential problems in relying on OLS estimates.12 First, the index of
literacy may be measured with substantial error. As we have argued before, many responses
11 By merging together the data on literacy, income, wealth and all the demographics needed for the empirical work, we end up with a final sample of 1,115 observations. 12 Note that we estimate a simple linear probability model. It is well-known that the error term of a linear probability model is heteroskedastic. Therefore, we correct the standard errors of the OLS estimates for the presence of heteroskedasticity. For the same reason, we use Generalized Method of Moments (GMM) estimation when we perform Instrumental Variables (IV) estimation.
69
Table 7. Multivariate analysis of stock market participation ____________________________________________________________________________________________________ (1)
OLS (2)
OLS (3)
OLS (4)
GMM (5)
GMM _________ _________ _________ _________ _________ Advanced literacy index 0.0839*** 0.0892*** 0.163** 0.155*** (0.012) (0.012) (0.069) (0.057) Basic literacy index 0.0112 -0.0138 (0.010) (0.023) Dummy (30<age<=40) -0.0250 -0.0101 -0.00850 0.00600 0.00384 (0.046) (0.045) (0.045) (0.048) (0.047) Dummy (40<age<=50) 0.0261 0.0326 0.0353 0.0474 0.0438 (0.047) (0.047) (0.047) (0.049) (0.048) Dummy (50<age<=60) 0.0133 0.0150 0.0165 0.0213 0.0195 (0.048) (0.047) (0.047) (0.048) (0.048) Dummy (age>60) 0.0604 0.0743 0.0734 0.0832 0.0841 (0.062) (0.060) (0.060) (0.059) (0.059) Intermediate vocational 0.0760* 0.0233 0.0247 0.0163 0.0148 (0.041) (0.036) (0.036) (0.037) (0.038) Secondary pre-university 0.0352 0.0249 0.0298 -0.0006 -0.0059 (0.037) (0.042) (0.041) (0.048) (0.051) Higher vocational 0.110*** 0.0676* 0.0717* 0.0471 0.0429 (0.037) (0.037) (0.037) (0.041) (0.044) University 0.153*** 0.0977** 0.102** 0.0691 0.0642 (0.047) (0.048) (0.047) (0.054) (0.057) Male 0.109*** 0.0715*** 0.0715*** 0.0428 0.0433 (0.028) (0.027) (0.027) (0.036) (0.035) Married -0.0367 -0.0280 -0.0267 -0.0167 -0.0184 (0.032) (0.031) (0.031) (0.032) (0.032) Number of children -0.00159 0.00371 0.00290 0.00538 0.00628 (0.015) (0.015) (0.015) (0.015) (0.015) Retired -0.0252 -0.0315 -0.0311 -0.0353 -0.0356 (0.055) (0.053) (0.053) (0.052) (0.052) Self-employed 0.0458 0.0315 0.0319 0.0232 0.0227 (0.056) (0.058) (0.057) (0.059) (0.059) Ln(household income) 0.0916*** 0.0845*** 0.0848*** 0.0790*** 0.0787*** (0.027) (0.026) (0.026) (0.027) (0.027) Second wealth quartile (€2300<wealth<=€45500) 0.100*** 0.0743** 0.0749** 0.0570 0.0568 (0.035) (0.035) (0.035) (0.039) (0.039) Third wealth quartile (€45500<wealth<=€197300) 0.155*** 0.117*** 0.117*** 0.0894** 0.0897** (0.037) (0.037) (0.037) (0.044) (0.044) Fourth wealth quartile (wealth>€197300) 0.212*** 0.159*** 0.160*** 0.122** 0.122** (0.042) (0.042) (0.042) (0.054) (0.054) Constant -0.886*** -0.752*** -0.760*** -0.664** -0.657** (0.26) (0.25) (0.25) (0.26) (0.26) Observations 1115 1115 1115 1115 1115 R-squared 0.10 0.14 0.14 0.11 0.12 Hansen J test p-value 0.673 0.672 F-statistic first stage regression 19.71 22.15 p-value exogeneity test 0.236 0.227 ____________________________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. This table reports OLS and GMM estimates of the effect of literacy on stock market participation. In the last two columns (GMM estimates), the advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
70
are imprecise and may result from simple guessing; this is particularly true for questions
measuring high levels of financial knowledge. Thus, OLS estimates may be biased
downward. On the other hand, there may also be learning and improvement in knowledge
(and familiarity with the questions asked in the module) via participation in the stock market.
This alternative argument leads to OLS estimates that are biased upward. In either case we
cannot simply rely on the estimates reported in the second and third column of Table 7 to
assess the effect of literacy.13
When we devised the module on financial literacy, we took into account the fact that
financial literacy is not an exogenous characteristic; in fact, literacy can itself be affected by
financial behavior (for example, if individuals learn via experience). To remedy this problem,
we have collected additional information (beyond current levels of economic knowledge) that
can serve as instruments for advanced financial literacy. To be able to rely on measures of
literacy that are exogenous with respect to stock market participation, we asked respondents
about their exposure to financial knowledge before entering the job market. Specifically, we
asked how much of their education was devoted to economics. 14 Note that economics is part
of the high school curriculum at the majority of schools in the Netherlands and it is possible
to specialize in economics/business at the high school level (economics degrees can be
pursued in college as well, of course).15 Our strategy is to rely on exposure to economic
education in the early stages of life. This measure should be correlated with current advanced
knowledge while it should be uncorrelated with stock market participation. As mentioned
before, advanced knowledge may be a crude proxy of actual knowledge. Moreover, it may
simply reflect how much respondents have learned from their personal experiences and from
their success in the stock market. For example, if financially knowledgeable respondents are
more likely to invest successfully and stay in the market, while low knowledge respondents
are more likely to lose money and exit the market, the relationship between literacy and
market participations may simply reflect the higher knowledge of those who stay in the
market.
13 The OLS estimates may also suffer from the omitted variables bias. For example, the error term may include ‘ability’ which is also correlated with financial literacy. As long as our measure of basic literacy index is a good proxy for ‘(financial) ability,’ we should not suffer from this problem. However, we address omitted variables bias later in the text. 14 For the precise wording of this question, see Appendix B. 15 In contrast to the US, there are no initiatives at the employer-level to improve financial literacy and economic knowledge of workers in the Netherlands. There are no retirement seminars, as the vast majority of Dutch employees participate in Defined Benefit retirement plans and have no say in their pension savings or the way their pension wealth is invested (see van Rooij, Kool and Prast (2007)). Thus, the supply of economic education is restricted to the school system in the Netherlands. Bernheim, Garrett and Maki (2001) show that those who were exposed to financial education in high school in the US were more likely to save later in life.
71
Table 8. First stage regressions _________________________________________________________________________________ (I) (II) _________________ _________________ Basic literacy index 0.290*** (0.027) Dummy (30<age<=40) -0.185* -0.166 (0.10) (0.12) Dummy (40<age<=50) -0.121 -0.0577 (0.099) (0.11) Dummy (50<age<=60) -0.0241 0.0155 (0.10) (0.12) Dummy (age>60) -0.0189 -0.0457 (0.13) (0.14) Intermediate vocational 0.0481 0.0943 (0.086) (0.095) Secondary pre-university 0.229*** 0.412*** (0.086) (0.090) Higher vocational 0.210*** 0.365*** (0.073) (0.077) University 0.357*** 0.555*** (0.080) (0.086) Male 0.299*** 0.345*** (0.058) (0.062) Married -0.119* -0.0988 (0.064) (0.068) Number of children -0.0247 -0.0534 (0.029) (0.033) Retired 0.0476 0.0656 (0.11) (0.11) Self-employed 0.119 0.151 (0.087) (0.10) Ln(household income) 0.0512 0.0703 (0.054) (0.057) Second wealth quartile (€2300<wealth<=€45500) 0.217** 0.269*** (0.093) (0.100) Third wealth quartile (€45500<wealth<=€197300) 0.342*** 0.409*** (0.090) (0.097) Fourth wealth quartile (wealth>€197300) 0.439*** 0.547*** (0.097) (0.10) Economics education: some -0.207*** -0.255*** (0.057) (0.064) Economics education: little -0.300*** -0.352*** (0.067) (0.073) Economics education: hardly at all or ‘don’t know’ -0.597*** -0.723*** (0.081) (0.092) Constant -0.642 -0.979* (0.53) (0.56) Observations 1115 1115 R-squared 0.33 0.22 p-value test age coefficients = 0 0.282 0.434 p-value test education coefficients = 0 0.000 0.000 p-value test wealth coefficients = 0 F statistic first stage regression
0.000 19.71
0.000 22.15
p-value test instruments =0 0.000 0.000 _________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
72
The first stage regressions are reported in Table 8. Responses to how much of
education was devoted to economics range from ‘hardly at all’ to ‘a lot’ and we construct
dummies for different levels of economics education while in school. These instruments have
a strong predictive power: Those who have had less exposure to economics education in
school are less likely to display advanced knowledge, and this holds true even when we
account for basic literacy, which we consider a measure of cognition and ability. The F-
statistic in the first stage regressions is high (with values close to 20) and beyond the values
recommended to avoid the weak instruments problem (Staiger and Stock, 1997; Bound,
Jaeger and Baker, 1995). The first stage results also continue to confirm the correlation
between literacy and demographic characteristics, such as education and gender, reported in
Table 4B.
The estimates in the second stage reported in the last two columns of Table 7 show
that the relationship between literacy and stock market participation remains positive,
statistically significant, and is even larger in the Generalized Method of Moments (GMM)
estimates. Moreover, the exogeneity test is not rejected. Thus, the OLS estimates do not differ
significantly from the GMM estimates. The results of the Hansen J-test show that the
overidentifying restrictions are not rejected. Overall, our estimates indicate that financial
literacy is an important determinant of stock market participation: Those who have low
financial knowledge are less likely to hold stocks.
6. Discussion and extensions
6.1 Exploiting stock market participation in the past
One of the potential objections concerning our instruments is that the exposure to
economics in school could be a choice variable, depending for example on tastes toward risk,
or perhaps simply reflecting ‘interest in the stock market’, i.e., how much respondents were
interested in becoming knowledgeable in economics to invest in the stock market. While this
may be the case for young generations, it can hardly be the case for middle-aged and older
respondents. Investing in the stock market is a recent phenomenon for many Dutch families
and it would be hard if not impossible for these families to have anticipated the current
changes in financial markets and the increase in individual responsibility.
To better understand and document household participation in the stock market, we
have examined other surveys that provide information about stock holdings in the 1980s. The
first wave of the Dutch Socio-Economic Panel, which covers a representative sample of the
73
population, shows that in 1987 only approximately 6% of families owned stocks (see also
Alessie, Lusardi and Aldershof (1997)), and that stock-ownership grew to only approximately
8% by 1990. Stock-ownership began to take off during the 1990s and it increased to more
than 20% by the end of the 1990s (Guiso, Haliassos and Jappelli, 2002). We exploit the
behavior of the stock market and the very recent increase in the fraction of families who own
stocks to further sharpen our understanding of the relationship between literacy and stock
market participation.
In Table 9A, we report the OLS and GMM estimates for respondents who are older
than 35. In this case, we concentrate on people who went to high school before 1990 during a
period when the stock market did not play any major role in the portfolios of most Dutch
families. Both the OLS and (most importantly) the GMM estimates remain positive and
statistically significant. Note that these estimates do not depend on the age split. We get
estimates of similar size when we split the sample at age 40 or at 45.
Table 9A. Stock market participation among respondents older than 35 _______________________________________________________________________________________ OLS OLS GMM GMM __________ __________ __________ __________ Advanced literacy index 0.0908*** 0.0964*** 0.146* 0.145** (0.015) (0.014) (0.066) (0.069) Basic literacy index 0.0136 -0.0015 (0.012) (0.025) Demographics (see table 7) yes yes yes yes Observations 884 884 884 884 R-squared 0.12 0.12 0.11 0.11 Hansen J test p-value 0.951 0.951 F-statistic first stage regression 18.97 20.11 p-value exogeneity test 0.476 0.466 _______________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group in the instrument set consists of those respondents whose education was devoted a lot to economics.
While it is admittedly hard to find good instruments for financial literacy, the
historical experience of the Netherlands provides us with a unique opportunity to rely on
information about financial literacy before the stock market became important and before
individuals took an active interest in the stock market. Since estimates of financial literacy do
not change significantly in size when considering respondents older than 35, in the next
sections we perform our estimates in the total sample.16
16 To further account for the fact that current or past literacy can proxy for ‘interest in economics’ we use answers to the question ‘How much understanding of economics do you need during daily activities (job,
74
To pursue this argument further and also investigate other instrument sets, we have
considered the information in the survey about advice from parents during childhood on how
to budget and save money in lieu of exposure to economics in school. However, we found no
relationship between this variable and advanced literacy. This provides further evidence that
the behavior of the stock market is a new experience and that current generations may be
unable to learn about investing in the stock market from previous generations. We turn next to
other potential sources of learning.
6.2 Stock market participation and peer effects
Another potential issue with the instruments we use is that respondents who were
exposed to economics during their schooling may be more likely to have friends (perhaps
their classmates) that invest in the stock market. Because of ‘peer effects’ in investing
respondents exposed to these friends may themselves be more likely to invest in the stock
market. Although we have previously documented that more financially knowledgeable
individuals are more likely to rely on formal sources of financial advice rather than relying on
family and friends, it is important to disentangle how much our variable measures ‘financial
knowledge’ versus ‘peer effects’. Several studies have documented that peer effects can be
pretty powerful determinants of portfolio choice (Hong, Kubik and Stein, 2004; Brown,
Ivkovic, Smith and Weisbenner, 2007) and those peer effects can start early in the life-cycle.
We have information in the data set on the level of education that most of the respondents’
acquaintances have. While this does not necessarily reflect knowledge of economics,
education is very strongly correlated with financial literacy as shown in Tables 4A and 4B.
In Table 9B, we report OLS and GMM estimates in a new empirical specification
where, in addition to the education of the respondents, we add the education of their peers (for
simplicity we only report the estimates of these new controls and the estimates for financial
literacy). The education level of peers does matter for stock-ownership. Those who have
friends that have a college degree are 12 to 14 percentage points more likely to own stocks.
Thus, there may be information-provision and learning via social interaction. Note, however,
that both the OLS and GMM estimates of literacy are barely affected by the addition of this
variable. Thus, financial literacy has an effect on stock ownership above and beyond the
effects of peers.
hobbies etc.)?’, available in the survey. Those who are not interested in economics are unlikely to choose a job that requires a lot of economics knowledge. Our measure of literacy continues to remain statistically significant at conventional levels even after the addition of dummies for the levels of ‘understanding of economics during daily activities’. For brevity, estimates are not reported but are available upon request.
75
Table 9B. Stock market participation and the importance of peer effects ________________________________________________________________________________ OLS OLS GMM GMM ________ ________ ________ ________ Advanced literacy index 0.0874*** 0.0930*** 0.158* 0.155** (0.012) (0.012) (0.086) (0.074) Basic literacy index 0.0145 -0.0039 (0.011) (0.024)
0.0748 0.0748 0.0539 0.0545 Education of peers: intermediate vocational, secondary pre-university (0.046) (0.046) (0.054) (0.054) Education of peers: higher vocational, university 0.143*** 0.144*** 0.119* 0.120* (0.055) (0.055) (0.064) (0.063) Demographics (see table 7) yes yes yes yes Observations 1054 1054 1054 1054 R-squared 0.16 0.16 0.14 0.14 p-value test education coefficients = 0 0.861 0.847 0.842 0.842 p-value test education peers coefficients = 0 0.030 0.029 0.102 0.101 Hansen J test p-value 0.842 0.840 F-statistic first stage regression 13.15 13.96 p-value exogeneity test 0.399 0.391 ________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group in the instrument set consists of those respondents whose education was devoted a lot to economics.
6.3 Self-assessed literacy versus objective literacy
Measuring literacy is clearly a difficult task. For example, we do not know how many
questions one should use to get a proper measure of literacy. Moreover, our questions are
focused on stocks and the stock market rather than financial knowledge in general. In this
section, rather than relying on our constructed indices, we use the simple measure of financial
literacy based on self-assessed economics knowledge. As mentioned before, we have asked
respondents to rate their understanding of economics on a scale from 1 to 7. This question is
easy to understand and to answer. Moreover, from a theoretical point of view, self-assessed
economics knowledge is what should influence household financial decision-making, even
though we show there is a strong correlation between subjective and objective measures of
knowledge. Finally, there is no mentioning of the stock market or financial market
instruments in this question and reverse causality may be less of a problem. On the other
hand, since the question refers to current economics knowledge, households may be
influenced in their judgment by their experience and success in the stock market. As before,
we first perform OLS regressions of stock market participation on financial literacy, this time
using self-assessed literacy in lieu of the literacy index. We then instrument self-assessed
knowledge, again using as instruments how much of the respondent education was devoted to
economics.
76
The estimates are reported in Table 9C.17 For brevity, we only report the estimates of
the variables of interest. Even when using this simple measure, the estimates of financial
literacy are positive and statistically significant. The GMM estimates are higher than the OLS
estimate and again the exogeneity test is not rejected. In both OLS and GMM regressions, we
account for the basic financial literacy index, which becomes statistically significant. Thus,
according to these alternative measures, both basic and self-assessed financial knowledge are
important determinants of stock market participation.
Table 9C. Stock market participation and self-assessed literacy _______________________________________________________________________ OLS GMM ____________ ____________ Self-assessed literacy 0.0629*** 0.0914** (0.012) (0.038) Basic literacy index 0.0332*** 0.0288** (0.011) (0.012) Demographics (see table 7) yes yes Observations 1083 1083 R-squared 0.13 0.13 Hansen J test p-value 0.624 F-statistic first stage regression 37.99 p-value exogeneity test 0.424 _______________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The self-assessed literacy question is reported in appendix B. The self-assessed literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
6.4 Knowledge or cognition?
One of the issues about financial literacy is whether it measures knowledge or simply
ability and cognition (see Benjamin, Brown and Shapiro (2006) and Stango and Zinman
(2007)). This distinction has important implications for public policy and, for example, for the
effectiveness of financial education programs. In our work, we try to account for cognition by
grouping together questions measuring the ability to perform simple calculations, the
understanding of changes in prices, and the time value of money (our basic literacy index).
We added this variable separately in the regressions in addition to the advanced knowledge
index. However, this is perhaps only a crude proxy of ability. To better account for cognition
and ability with calculations, we exploited two important economic changes in the
Netherlands. First, like most of the members of the European Union, the Netherlands shifted
17 In the regression analysis, we deleted the respondents who did not know the answer to this question or refused to answer.
77
from their national currency (the Dutch guilder) to the Euro. As of 2002, the Euro replaced
the guilder as a legal mean of payment. We exploited this fact in the second module that was
added to the DNB survey in January 2006. We asked respondents how difficult it was to do
shopping, read bank statements, and do typical daily transactions right after the introduction
of the Euro in 2002 (answers range from ‘very difficult’ to ‘not difficult at all’).18 More than
13% of respondents found the conversion to the Euro to be ‘very difficult’ or ‘difficult’,
21.9% found it ‘somewhat difficult’ and the rest (63%) found it ‘not very difficult’ or ‘not
difficult at all’. We constructed dummies for the responses to this question and added them to
the regression to account for cognitive ability (these dummies replaced the basic financial
literacy index). When we account for these dummies in our regressions, both the OLS and the
GMM estimates of the advanced literacy index remain positive, statistically significant and of
similar magnitude. Thus, financial literacy affects stock ownership above and beyond the
effect of cognition and the ability to perform calculations.
We also considered another important change in the Netherlands, this time concerning
the health system. A new law was passed in 2005 that introduced more freedom of choice in
the health insurance system. Households were required to make decisions about their health
providers, their contributions, and the deductible in their health policy. Decisions had to be
made before March 1, 2006 (the ultimate deadline to make changes to previous decisions at
no cost). In the new module we added in January 2006, we ask respondents how difficult it
was to understand the new health insurance system (again, answers can range from ‘very
difficult’ to ‘not difficult at all’). 19 However, contrary to the conversion to the Euro - where
respondents were confronted with a currency exchange and had to make simple calculations -
there are several reasons why the new health system is difficult to comprehend.20 We further
asked respondents the reasons for their answer, in order to differentiate between those who
did not know how to make this kind of decision (low cognitive ability respondents), and those
who considered the decision difficult because they had to spend time reading and collecting
information and had to figure out what was best for them to do (high cognitive ability
respondents).
Overall, 43% of respondents found the health decisions ‘not very difficult’ or ‘not
difficult at all’. Of the remaining group who found the decision ‘very difficult’, ‘difficult’ or
‘somewhat difficult’, more than half reported that it was because they had to spend time to
18 For the precise wording of this question, see Appendix B. 19 For the precise wording of these questions, see Appendix B. 20 People had to choose from a large number of health insurers and had to compare the coverage and price of supplementary health packages, which offered different deductibles.
78
make comparisons and reading and collecting information. As before, we constructed
dummies for different types of respondents and added these dummies to our regression. Even
after controlling for this alternative measure of cognitive ability, we find that both the OLS
and GMM estimates of the advanced literacy index remain positive and statistically
significant (Table 9D).
Table 9D. Stock market participation and alternative measures of basic literacy __________________________________________________________________________________________ Euro Introduction Change Health
Insurance System ___________________ ____________________ OLS GMM OLS GMM ________ ________ ________ ________ Advanced literacy index 0.0848*** 0.141** 0.0880*** 0.156** (0.012) (0.061) (0.012) (0.065) Dealing with Euro: somewhat difficult -0.0469 -0.0521 (0.045) (0.046) Dealing with Euro: not very difficult -0.0138 -0.0240 (0.042) (0.044) Dealing with Euro: not difficult at all 0.0450 0.0289 (0.048) (0.052)
-0.0105 -0.00622 Difficulty health system: making comparisons and collecting info (0.030) (0.031) -0.0257 -0.00807 Difficulty health system: figuring out what the best for me to do (0.036) (0.040) 0.0755 0.131 Difficulty health system: I don’t know how to make these
decisions & DK (0.075) (0.088) Demographics (see table 7) yes yes yes yes Observations 1053 1053 1053 1053 R-squared 0.14 0.13 0.14 0.12 p-value test Euro coefficients = 0 0.156 0.236 p-value test health insurance coefficients = 0 0.590 0.398 Hansen J test p-value 0.960 0.970 F-statistic first stage regression 18.37 17.26 p-value exogeneity test 0.343 0.280 __________________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. In the first two columns, the reference group consists of those respondents who found dealing with the Euro transition ‘very difficult’ or who answered the question with ‘do not know’. In the last two columns, the reference group consists of those respondents who have no difficulty understanding the health care system change (see question H1 in appendix B). The three dummy variables are based on question H2 in appendix B. The advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
6.5 A different financial literacy index
As mentioned before, to assess the quality of the answers to literacy questions, we
changed the wording of three questions and exposed two randomly selected groups of
respondents to the same question with different wording. From this methodology we inferred
that respondents had considerable difficulty understanding the questions about bond pricing
79
and the riskiness of a company stock versus a stock mutual fund. In performing the factor
analysis, respondents were divided into different subgroups according to the wording of the
question they were exposed to. Since there may be a lot of noise in the answers to these
questions, in this section we perform the empirical analysis excluding the three questions for
which we implemented a different wording.21 In this way, we can show how sensitive our
estimates are not only to our methodology, but also to different measures of literacy. By
excluding these questions, we exclude concepts that were rather difficult for respondents to
grasp, and we can therefore check whether indices that have a stronger focus on basic
economic concepts are still related to stock ownership.
As in the previous tables, we report both OLS and GMM estimates. Since we exclude
questions explicitly related to stocks and the pricing of bonds, the problem of reverse
causality may be less prevalent. At the same time, we may have decreased the amount of
noise in the index, since it is hard to infer a lot from answers related to topics that respondents
do not know well. The OLS estimates in Table 9E shows that literacy is still related to stock
market participation, even when we focus on an index that excludes several advanced
economic concepts. The GMM estimates are also positive and statistically significant and of
similar magnitude than the previous estimates.
Table 9E. Stock market participation and an alternative advanced literacy index __________________________________________________________________________________________ OLS OLS GMM GMM __________ __________ __________ __________ Advanced literacy index (alternative) 0.0767*** 0.0823*** 0.182** 0.166*** (0.012) (0.012) (0.078) (0.062) Basic literacy index 0.0113 -0.0243 (0.010) (0.028) Demographics (see table 7) yes yes yes yes Observations 1115 1115 1115 1115 R-squared 0.13 0.13 0.09 0.10 Hansen J test p-value 0.684 0.682 F-statistic first stage regression 16.15 19.07 p-value exogeneity test 0.163 0.156 __________________________________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The (alternative) advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
We have also experimented with excluding questions 12 and 13 from the set of
advanced literacy questions since the latter has a very low correct response rate and there is
21 See Appendix A for the calculation of the financial literacy index.
80
already one question in the set about risk diversification. In addition, we experimented with
excluding questions 7 and 9, which simply refer to the definition of stocks and bonds.
Estimates for financial literacy remain positive and statistically significant. For example, the
GMM estimates are 0.159 (s.e. 0.067) and 0.174 (s.e. 0.074) in the first and second case
respectively. Thus, results do not depend on the inclusion or exclusion of a particular question
in the literacy index.
6.6 Including measures of risk aversion
Notably, one of the variables which is missing from our empirical specification is a
measure of risk aversion. Clearly, preferences for risk are an important determinant of stock
ownership and may explain some of the differences among households.22 Some researchers
have further argued that knowledge and cognitive ability may have an effect on preferences,
such as risk aversion and the rate of time preference (Benjamin, Brown and Shapiro, 2006;
Dohmen, Falk, Huffman and Sunde, 2007) and, through this channel, affect financial
decision-making. We do not investigate this relationship in our paper, but will account for
preferences in a new empirical specification. In this way, our indices can better measure the
effects of knowledge and information costs rather than the effect of preferences. In a separate
module on preferences in the DHS, there are questions that aim to measure attitudes toward
risk. These questions are similar to those in the HRS.23 Barsky, Juster, Kimball and Shapiro
(1997) show that, while imperfect, the measure of risk aversion derived from these types of
questions is related to financial behavior and correlates with stock ownership. However, one
of the disadvantages of using the risk aversion data is that we lose a lot of observations from
merging together separate sections of DHS.
From the information provided in the survey, we can distinguish among four types of
households, from those unwilling to take any risk (reject any gamble that offers higher but
uncertain payoff) to those willing to take substantial risk (willing to take both gambles
presented in the questions that offer high but uncertain payoffs). When we examine a simple
correlation between stock market participation and our risk aversion dummies, we find that
risk is correlated to ownership of stocks: Those who are not willing to take risk are less likely
to participate in the stock market. Thus, while a crude measure, the risk aversion dummies
seem to be able to proxy for attitudes toward risk.
22 However, as reviewed in Haliassos and Bertaut (1995), risk aversion alone cannot explain why so many households do not hold stocks. One has to appeal to different preferences than the general class of HARA preferences to explain lack of stockownership. 23 For the precise wording of these questions, see appendix B.
81
When including risk aversion in our empirical specification in Table 9F, we find that
the estimates of our variables of interest do not change. Both the OLS and GMM estimates of
financial literacy remain positive, statistically significant, and do not change appreciably in
magnitude. Thus, the exclusion of risk aversion does not take away from the importance of
financial literacy in explaining participation in the stock market.
Table 9F. Stock market participation, literacy, and risk aversion _______________________________________________________________________ OLS GMM ______________ ______________ Advanced literacy index 0.0974*** 0.151* (0.014) (0.080) Basic literacy index 0.00477 -0.0112 (0.012) (0.026) Risk aversion: low -0.0431 -0.0627 (0.084) (0.094) Risk aversion: medium 0.0172 -0.00714 (0.055) (0.066) Risk aversion: high 0.0558 0.0451 (0.045) (0.047) Risk aversion: don’t know 0.0185 0.0344 (0.063) (0.068) Demographics (see table 7) yes yes Observations 888 888 R-squared 0.13 0.12 Hansen J test p-value 0.480 F-statistic first stage regression 15.48 p-value exogeneity test 0.493 _______________________________________________________________________ Note: Robust standard errors in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The advanced literacy index has been instrumented using three dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics. In this regression the reference group consist of those respondents who exhibit the highest degree of risk aversion according to the questions reported in appendix B.
6.7 Other extensions
We have pursued another robustness check to show that financial literacy is an
important determinant of stock-ownership and captures information and search costs related
to a complex asset such as stocks. In addition to stocks, we have examined the relationship
between financial literacy and savings accounts. A much lower degree of financial
sophistication and information costs is required to deal with these assets and we would not
expect to find a strong relationship with financial literacy. Indeed, in our empirical work, we
do not find any relationship between our measures of literacy and ownership of savings
accounts. The OLS and GMM estimates of advanced literacy are 0.0167 (s.e. 0.014) and
82
0.0142 (s.e. 0.059) respectively. This confirms the results of Christelis, Jappelli and Padula
(2007), who also found no relationship between cognitive ability and savings accounts.
Our results are robust to a variety of other specifications. For example, we have
excluded from our sample respondents who are older than 70, which should be in the
decumulation phase of their life-cycle. This increases the power of our instruments, since the
effect of schooling declines with age. The OLS and GMM estimates of advanced literacy are
0.082 (s.e. 0.013) and 0.167 (s.e. 0.071) respectively. Moreover, rather than simply
accounting for self-employment in our specification, we have excluded the self-employed
from our sample. Hurst and Lusardi (2007) show that the self-employed/business owners
display many differences with respect to other households and we do not have a lot of
information in our data set to account for all these differences. However, our OLS estimate of
financial literacy is 0.088 (s.e. 0.012) and the GMM estimate is 0.138 (s.e. 0.068). Thus,
estimates continue to remain positive and statistically significant
7. Concluding remarks
In this paper, we show that lack of understanding of economics and finance is a
significant deterrent to stock ownership. The different measures of financial knowledge we
have employed in our work all show that lack of literacy prevents households from
participating in the stock market. Cocco, Gomez and Maenhout (2005) show that the welfare
loss from non-participation in the stock market can be sizable. Thus, the role of financial
literacy should not be under-estimated. As more workers transition to a system where they
have to decide how much to save for retirement and how to invest their retirement wealth, it is
important to consider ways to enhance their level of financial knowledge or to guide them in
their financial decisions.
We plan to expand this work in several directions. First, we will examine the
relationship between financial literacy and retirement planning and explore whether
difficulties in performing calculations and low financial sophistication affect also the ability
to plan for retirement. Moreover, we will assess whether financial literacy has an effect not
only on portfolio choice but also on savings behavior and whether those who display low
literacy are less likely to accumulate wealth.
83
References Agnew, J., and L. Szykman, 2005, Asset allocation and information overload: The influence of information display, asset choice and investor experience, Journal of Behavioral Finance, 6, 57-70. Agarwal, S., J. Driscoll, X. Gabaix, and D. Laibson, 2007, The age of reason: Financial decisions over the lifecycle, mimeo, Harvard University. Alessie, R., S. Hochguertel, and A. Van Soest, 2002, Household portfolios in the Netherlands, In: L. Guiso, M. Haliassos, and T. Jappelli (eds.), Household Portfolios, MIT Press, Cambridge. Alessie, R., A. Lusardi, and T. Aldershof, 1997, Income and wealth over the life cycle: Evidence from panel data, Review of Income and Wealth, 43, 1-32. Barsky, R., F. Juster, M. Kimball, and M. Shapiro, 1997, Preference parameters and behavioral heterogeneity: An experimental study in the Health and Retirement Study, Quarterly Journal of Economics, 112, 537-579. Bartlett, M., 1937, The statistical conception of mental factors, British Journal of Psychology, 28, 97-104. Benjamin D., S. Brown, and J. Shapiro, 2006, Who is ‘behavioral’?: Cognitive ability and anomalous preferences, mimeo, Harvard University. Bernheim, D., 1995, Do households appreciate their financial vulnerabilities? An analysis of actions, perceptions, and public policy. In: Tax Policy and Economic Growth, American Council for Capital Formation, Washington, DC, 1-30. Bernheim, D., 1998, Financial illiteracy, education and retirement saving. In: O. Mitchell and S. Schieber (eds.), Living with Defined Contribution Pensions, University of Pennsylvania Press, Philadelphia, 38-68. Bernheim, D., D. Garrett, and D. Maki, 2001, Education and saving: The long-term effects of high school financial curriculum mandates, Journal of Public Economics, 80, 435-565. Bernheim, D., and D. Garrett, 2003, The effects of financial education in the workplace: Evidence from a survey of households, Journal of Public Economics, 87, 1487-1519. Bound, J., D. Jaeger, and R. Baker, 1995, Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak, Journal of the American Statistical Association, 90, 443-450. Brown, J., Z. Ivkovic, P. Smith, and S. Weisbenner, 2007, Neighbors matter: Causal community effects and stock market participation, NBER Working Paper, 13168. Campbell, J., 2006, Household finance, Journal of Finance, 61, 1553-1604.
84
Chang, L., and J. Krosnick, 2003, National surveys via RDD telephone interviewing versus the Internet: Comparing sample representativeness and response quality, mimeo, Ohio State University. Christelis, D., T. Jappelli, and M. Padula, 2007, Cognitive abilities and portfolio choice, mimeo, University of Salerno. Cocco, J., F. Gomes, and P. Maenhout, 2005, Consumption and portfolio choice over the life-cycle, Review of Financial Studies, 18, 490-533. Constantinides, G., J. Donaldson, and R. Mehra, 2002, Junior can’t borrow: A new perspective on the equity premium puzzle, Quarterly Journal of Economics, 117, 269-296. Davis, S., F. Kubler, and P. Willen, 2006, Borrowing costs and the demand for equity over the life cycle, Review of Economics and Statistics, 88, 348-362. Dohmen, T., A. Falk, D. Huffman, and U. Sunde, 2007, Are risk aversion and impatience related to cognitive ability?, IZA Discussion Paper, 2735. Guiso L., M. Haliassos, and T. Jappelli, 2002, Household portfolios, MIT Press, Cambridge. Guiso, L., and T. Jappelli, 2005, Awareness and stock market participation, Review of Finance, 9, 537-567. Guiso, L., P. Sapienza, and L. Zingales, 2005, Trusting the stock market, NBER Working Paper, 11648. Haliassos, M., and C. Bertaut, 1995, Why do so few hold stocks?, Economic Journal, 105, 1110-1129. Hilgert, M., and J. Hogarth, 2002, Financial knowledge, experience and learning preferences: Preliminary results from a new survey on financial literacy, Consumer Interest Annual, 48. Hilgert, M., J. Hogarth, and S. Beverly, 2003, Household financial management: The connection between knowledge and behavior, Federal Reserve Bulletin, 309-322. Heaton, J., and D. Lucas, 2000, Portfolio Choice and asset prices: The importance of entrepreneurial risk, Journal of Finance, 55, 1163-1198. Hong, H., J. Kubik, and J. Stein, 2004, Social interaction and stock-market participation, Journal of Finance, 59, 137–163. Hurd, M., 1990, Research on the elderly: Economic status, retirement and consumption and saving, Journal of Economic Literature, 28, 565-637. Hurst, E., and A. Lusardi, 2007, Liquidity constraints and entrepreneurship. Household wealth, parental wealth, and the transition in and out of entrepreneurship, forthcoming in Overcoming Barriers to Entrepreneurship, Rowman and Littlefield.
85
John Hancock Financial Services, 2002, Insight into participant investment, knowledge and behavior, Eighth Defined Contribution Survey. Kimball, M., and T. Shumway, 2006, Investor sophistication and the participation, home bias, diversification, and employer stock puzzles, mimeo, University of Michigan. Lusardi, A., and O. Mitchell, 2006, Financial literacy and planning: Implications for retirement wellbeing, Pension Research Council Working Paper, 1, The Wharton School. Lusardi, A., and O. Mitchell, 2007a, Baby boomers retirement security: The role of planning, financial literacy and housing wealth, Journal of Monetary Economics, 54, 205-224. Lusardi, A., and O. Mitchell, 2007b, Financial literacy and retirement preparedness: Evidence and implications for financial education, Business Economics, 35-44. Mandell, L, 2004, Financial literacy: Are we improving? Washington, D.C.: Jump$tart Coalition for Personal Financial Literacy. Moore, D., 2003, Survey of financial literacy in Washington State: Knowledge, behavior, attitudes and experiences, Technical report,03-39, Social and Economic Sciences Research Center, Washington State University. National Council on Economic Education, 2005, What American teens and adults know about economics, Washington, D.C. Nyhus, E., 1996, The VSB-CentER saving project: Data collection methods, questionnaire and sampling procedures, Progress Report 42, Tilburg University. Organization for Economics Co-Operation and Development, 2005, Improving financial literacy: Analysis of issues and policies, Paris, France. Stango, V., and J. Zinman, 2007, Fuzzy math and red ink: When the opportunity cost of consumption is not what it seems, Working Paper, Dartmouth College. Staiger, D., and J. Stock, 1997, Instrumental variables regression with weak instruments, Econometrica, 65, 557-586. Van Rooij, M., C. Kool, and H. Prast, 2007, Risk-return preferences in the pension domain: Are people able to choose?, Journal of Public Economics, 91, 701-722.
86
Appendix A. Constructing indices for basic and advanced financial literacy The index for basic literacy is based on the first 5 questions reported in Section 4. For each basic
literacy question we have constructed a dummy variable for respondents who answered correctly to
the question. We have performed a factor analysis on those binary variables using the iterated
principal factor method. We were able to retain one factor with a meaningful interpretation; this factor
describes basic literacy. The factor loadings are presented in Table A1. Given these factor loadings,
we obtained factor scores using the Bartlett method (Bartlett, 1937).
Table A1. Factor loadings corresponding to the five basic literacy questions _________________________________________ Basic literacy questions Factor loadings _____________________ _________________ Numeracy 0.6667 Interest compounding 0.5188 Inflation 0.5513 Time value of Money 0.4267 Money illusion 0.2432 _________________________________________ The advanced financial literacy index has been constructed using the next 11 questions presented in
Section 4. As we state in the main text, three questions were ‘randomized’ (see Table 3). The
following two items presented in Table 3 are very sensitive to the way the question is formulated.
(15a) Buying a company stock usually provides a safer return than a stock mutual fund? (15b) Buying a stock mutual fund usually provides a safer return than a company stock? (16a) If the interest rate falls, what should happen to bond prices: rise/fall/stay the same/none of the
above? (16b) If the interest rate rises, what should happen to bond prices: rise/fall/stay the same/none of the
above? Therefore, we decided to split the sample into four groups and to perform the factor analysis on each
of those four groups separately. The first group had to answer questions 15a and 16a, the second group
15b and 16a, the third group 15a and 16b and the fourth group 15b and 16b. Since the assignment to
those groups occurred randomly with equal probability (25%), the sub-samples are about of equal size.
Contrary to the answers to the basic literacy questions, the responses to the advanced literacy
questions include many ‘do not know’ answers. To take this response behavior into account, we
constructed 2 dummy variables for each of the 11 questions. The first dummy variable indicates
whether the question was answered correctly, while the other one refers to the ‘do not know’ answers.
In other words, we performed a factor analysis on 22 variables. We were able to retain one factor with
a meaningful interpretation: it basically describes advanced literacy. The factor loadings are presented
in Table A2.
87
Table A2. Factor loadings for the advanced literacy questions (four subsamples) ________________________________________________________________________________________ Advanced literacy questions Factor loadings ____________________________________________ _________________________________ 15a,16a 15b,16a 15a,16b 15b,16b ______ ______ ______ ______
Correct 0,3602 0,3903 0,3548 0,3819 If the interest rate falls, what should happen to bond prices: rise/fall/stay the same/none of the above? DK -0,6607 -0,7346 -0,6863 -0,7072
Correct 0,6787 0,441 0,6512 0,4177 Buying a company stock usually provides a safer return than a stock mutual fund? DK -0,7688 -0,8016 -0,7554 -0,7158
Correct 0,5883 0,6798 0,6036 0,6196 Stocks are normally riskier than bonds? DK -0,7257 -0,819 -0,7194 -0,7786
Correct 0,4684 0,5099 0,5549 0,5293 Considering a long time period, which asset described below normally gives the highest return: Savings accounts, Bonds or Stocks?
DK -0,6964 -0,7655 -0,7993 -0,7245
Correct 0,6459 0,6731 0,6532 0,6655 Normally, which asset described below display the highest fluctuations over time: Savings accounts, Bonds or Stocks?
DK -0,7548 -0,7904 -0,7954 -0,7516
Correct 0,4980 0,5804 0,5578 0,6159 When an investor spreads his money among different assets, does the risk of losing money increase, decrease or stay the same?
DK -0,7410 -0,7685 -0,7441 -0,7532
Correct 0,4798 0,4658 0,4669 0,5176 If you buy a 10-year bond, it means you cannot sell it after 5 years without incurring a major penalty. True or false?
DK -0,6373 -0,6398 -0,6414 -0,6652
Correct 0,5646 0,6848 0,5584 0,6003 Which of the following statements describes the main function of the stock market? 1) DK -0,7178 -0,7457 -0,6948 -0,7190
Correct 0,4489 0,4619 0,3862 0,4452 What happens if somebody buys the stock of firm B in the stock market? 1) DK -0,6619 -0,6764 -0,6227 -0,5875
Correct 0,5931 0,6754 0,6331 0,6479 Which statement about mutual funds is correct? 1)
DK -0,7507 -0,7925 -0,7816 -0,7253 Correct 0,5829 0,6365 0,5852 0,6436 What happens if somebody buys a bond of firm B? 1)
DK -0,7178 -0,8032 -0,7434 -0,7402 ________________________________________________________________________________________
1) See the exact wording of the question in the Box 2.
88
We have also constructed an alternative index for advanced financial literacy where we do not use the
questions that were randomized (see Table 3). The results of the factor analysis (factor loadings) are
shown in Table A3. This alternative index has been used in the sensitivity analysis presented in Table
9E.
Table A3. Factor loadings for the advanced literacy questions excluding the randomized questions __________________________________________________________________________________________ Advanced literacy questions (excluding the three randomized questions) Factor
loadings _____________________________________________________________________ _________
Correct 0,5166 Considering a long time period, which asset described below normally gives the highest return: Savings accounts, Bonds or Stocks? DK -0,7527
Correct 0,6522 Normally, which asset described below display the highest fluctuations over time: Savings accounts, Bonds or Stocks? DK -0,7874
Correct 0,5820 When an investor spreads his money among different assets, does the risk of losing money increase, decrease or stay the same? DK -0,7682
Correct 0,4545 If you buy a 10-year bond, it means you cannot sell it after 5 years without incurring a major penalty. True or false? DK -0,6175
Correct 0,6292 Which of the following statements describes the main function of the stock market? 1)
DK -0,7443 Correct 0,4408 What happens if somebody buys the stock of firm B in the stock market? 1)
DK -0,6615 Correct 0,6521 Which statement about mutual funds is correct? 1)
DK -0,7704 Correct 0,5975 What happens if somebody buys a bond of firm B? 1)
DK -0,7372 __________________________________________________________________________________________ 1) See the exact wording of the question in Box 2.
89
Appendix B. Exact wording of the questions in the questionnaire and construction of variables used in the empirical work. Self-assessed literacy How would you assess your understanding of economics (on a 7-point scale; 1 means very low and 7 means very high)?
Very low Very high [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7
[ ] Do not know [ ] Refusal
The index of self-assessed literacy used in the regression analysis is constructed by grouping together the two lowest categories (very few respondents have chosen the lowest level), recoding the remaining six levels of self-assessed literacy from 1 to 6 and excluding ‘do not know’ answers and ‘refusals.’ Economics education How much of your education was devoted to economics? [ ] A lot [ ] Some [ ] Little [ ] Hardly at all [ ] Do not know [ ] Refusal The instrument variable economics education in the past is used in the regression analysis by including three dummy variables for the response categories ‘some’, ‘little’ and ‘hardly at all,’ respectively. The reference group consists of those respondents whose education was devoted ‘a lot’ to economics. The ‘do not knows’ and ‘refusals’ are grouped together with the ‘hardly at all’ answers. Conversion to Euro In 2002 we went from the guilder to the Euro. How difficult was it for you back then to go shopping, read your bank statements and do your usual daily transactions using the Euro? [ ] Very difficult [ ] Difficult [ ] Somewhat difficult [ ] Not very difficult [ ] Not difficult at all [ ] Do not know [ ] Refusal The variable conversion to Euro is used in the regression analysis by including three dummy variables for the response categories ‘somewhat difficult’, ‘not very difficult’ and ‘not difficult at all,’ respectively. The reference group consists of those respondents who found the transition from the guilder to the Euro ‘very difficult’ or ‘difficult’. ‘Do not knows’ and ‘refusals’ are grouped together with these latter two categories. Health care system change H1) This year, the Dutch system of health insurance has changed. How difficult is it for you to understand the new Health Insurance system? [ ] Very difficult [ ] Difficult
90
[ ] Somewhat difficult [ ] Not very difficult [ ] Not difficult at all [ ] Do not know [ ] Refusal [If the response to question H1 is not equal to ‘not very difficult’ or ‘not difficult at all’ then the following question (H2) is asked] H2) Could you please indicate which of the following statements best describes what makes the decisions you have to make difficult? [ ] I have to make comparison and spend time reading and collecting information [ ] I have to find a way to figure out what is best for me to do [ ] I do not know how to make this kind of decisions [ ] Do not know [ ] Refusal The variable health care system change is used in the regression analysis by including three dummy variables for the first three response categories in question H2. The ‘do not know’ and ‘refusal’ answers are grouped together with the group which indicated ‘I do not know how to make this kind of decisions’. The reference group consists of those respondents who reported they find the change in the system of health insurance either ‘not very difficult’ or ‘not difficult at all.’ Risk aversion R1) Suppose that you are the only income earner in the family, and you have a good job guaranteed to give you your current (family) income every year for life. You are given the opportunity to take a new, equally good job, with a 50% chance it will double your (family) income and a 50% chance that it will cut your (family) income by a third. Would you take the new job?
[ ] Yes [ ] No [ ] Do not know [If R1=‘yes’ then R2] R2) Suppose the chances were 50% that it would double your (family) income, and 50% that it would cut it in half. Would you take the new job?
[ ] Yes [ ] No [ ] Do not know [If R1=‘no’ or ‘do not know’ then R3] R3) Suppose the chances were 50% that it would double your (family) income and 50% that it would cut it by 20 percent. Would you then take the new job? [ ] Yes [ ] No [ ] Do not know The variable risk aversion is used in the regression analysis by including four dummy variables: One for those who choose the most risky option twice (least risk averse), one for those who choose the most risky option the first question but not in the second question (medium risk averse), one for those who choose the safe option in the first question but not in the second question (risk averse) and one for those who do not make a choice in the first question (do not know), respectively. The reference group consists of those respondents who choose the safe option twice (most risk averse).
91
PAPER THREE
FINANCIAL LITERACY, RETIREMENT PLANNING, AND HOUSEHOLD WEALTH
(Co-authors: Annamaria Lusardi and Rob Alessie)
93
Financial Literacy, Retirement Planning, and Household Wealth
Maarten van Rooij (De Nederlandsche Bank and Netspar) Annamaria Lusardi (Dartmouth College and NBER)
Rob Alessie (University of Groningen, Netspar and Tinbergen Institute)*
November 2008
Abstract
There is ample empirical evidence documenting widespread financial illiteracy and limited pension knowledge. At the same time, the wealth distribution is heavily dispersed and many workers arrive on the verge of retirement with little or no personal assets. This paper is the first to investigate the relation between financial sophistication and household net worth relying on specific measures of financial knowledge and skills rather than crude proxies. For this purpose, we have designed a new module for the Dutch DNB Household Survey. Our findings provide evidence of a statistically and economically significant positive effect of financial sophistication on net worth. Moreover, we highlight empirical evidence of two channels by which financial sophistication facilitates wealth accumulation. First, financial skills increase the likelihood to invest in the stock market thereby opening the possibility to benefit from the equity premium and improving the opportunities for risk diversification. Second, financial sophistication boosts retirement planning behavior by households, thereby providing an important channel for the development of savings plans and creating instruments for self-control. In addition, our results suggest that respondents who are relatively confident on their own financial skills have a higher propensity to plan. To take into account that wealth, portfolio management and planning activities might exert an independent effect on financial literacy, we employ instrumental variable regression techniques using information on economics education. Key words: Financial Education, Savings and Wealth Accumulation, Retirement Planning, Knowledge of Finance and Economics, Overconfidence, Stock Market Participation JEL Classification: D91, D12, J26 ∗ Maarten C.J. van Rooij, Economics & Research Division, De Nederlandsche Bank, P.O. Box 98, 1000 AB, Amsterdam ([email protected]), Annamaria Lusardi, Department of Economics, Dartmouth College, Hanover, NH 03755 ([email protected]), and Rob J.M. Alessie, School of Economics and Business, University of Groningen, P.O. Box 800, 9700 AV, Groningen ([email protected]). We gratefully acknowledge valuable comments and suggestions by Carol Bertaut, Johannes Binswanger, Lex Hoogduin, Clemens Kool, Peter Schotman, Federica Teppa, Peter van Els, and participants in the Netspar pension conference (The Hague, June 2008), and the ECB-CFS conference on Household Finances and Consumption (Frankfurt, September 2008). We thank the staff of CentERdata and, in particular, Corrie Vis for their assistance in setting up the survey and the field work. The views expressed in this paper are those of the authors and do not necessarily reflect official positions of De Nederlandsche Bank.
94
1. Introduction
Households hold very different amounts of savings. Heterogeneity in lifetime
earnings, the willingness to leave bequests, motives for precautionary or other savings, and
cross sectional variability in time preferences, expectations, health, longevity, inheritances
and other income shocks contribute to the dispersion in wealth holdings and have been
researched extensively.1 The relation between wealth accumulation and financial capabilities
has received much less attention, mainly because information on the level of financial
sophistication is usually unavailable. Recently, however, there has been a boost in research on
the measurement of financial literacy and its effects on household behavior (e.g. Van Rooij,
Lusardi and Alessie, 2007; Lusardi and Mitchell, 2007a, 2007b, 2008; Agnew, Szykman,
Utkus and Young, 2007; Kimball and Shumway, 2006). In this paper, we report the results of
a new survey with an extensive set of questions designed to measure basic and more advanced
financial skills and to the best of our knowledge it is the first study of its impact on net worth.
The relation between financial sophistication and savings behavior is important as
individuals are increasingly asked to take private responsibility for their financial well-being.
Given the evidence on widespread financial illiteracy and limited pension knowledge, there is
an obvious policy interest in the question whether financial education affects savings behavior
and what type of education programs is most effective. The empirical evidence on the effect
of financial education and the provision of information on savings behavior is mixed (Lusardi,
2004). Moreover, even if studies find a significant impact of financial education on savings,
the outcomes generally do not provide much information on the channel underlying this
effect. Studies on the impact of retirement seminars for example are typically not able to
disentangle the consequences of an increase in financial skills, if any, from behavioral effects
due to the provision of information, retirement seminars being an integral part of a more
comprehensive initiative to increase financial awareness, or the importance of peer and
community effects in raising savings (Duflo and Saez, 2003). We isolate the effect of
financial skills, investigate whether financial sophistication as such has an impact on wealth
accumulation and ask ourselves what underlying channels are at work here.
The main contributions of this paper are the following. We provide evidence of an
independent and positive effect of financial sophistication on wealth accumulation over and
above the effect of other determinants such as income, age, education, family composition,
risk tolerance, patience, the attitude towards saving, and basic cognitive ability. We identify
and highlight two channels by which financial literacy facilitates wealth accumulation. First, a 1 See the references in the next section.
95
high level of financial skills lowers the costs of gathering and processing information and
reduces barriers to invest in the stock market (Haliassos and Bertaut, 1995; Vissing-
Jorgenson, 2004). We show that financial sophistication indeed fosters stock market
participation and thereby provides households with the opportunity to benefit from the equity
premium on stock investments. Second, we find that financial sophistication boosts retirement
planning behavior by households, thereby providing an important mechanism for wealth
accumulation (Ameriks, Caplin and Leahy, 2003; Lusardi and Mitchell, 2007a). In addition,
our empirical results suggest that respondents who are relatively confident about their own
financial skills have a higher propensity to plan. The intuition behind the retirement planning
channel is that a high level of financial knowledge and skills reduces planning costs, i.e. the
economic and psychological barriers to acquire information, to do the calculations and to
develop a plan. Our data show that once households start doing calculations on their savings
needs for retirement, they often follow through setting up a retirement plan and are in general
also successful in sticking to their plan.
Our findings have important policy implications. Financial skills cannot be taken for
granted. We show that financial illiteracy is widespread and that the lack of financial
sophistication has important consequences for wealth holdings. This suggests that the skills to
take financial decisions often fall short of what is necessary for the kind of choices that
individuals nowadays are expected to make in a financial world with a vast and growing
supply of complicated products which have become accessible to a large public by now. The
implication is that there is an important role for financial education as by effectively boosting
financial sophistication households become better equipped to manage their own savings. One
reason why this is important is that many households enter retirement with very little wealth
(Venti and Wise, 1998, 2000; Lusardi, 1999, 2003). This has profound implications not only
for personal welfare but also for public policy, as low savings households lack a buffer to deal
with negative shocks and are more likely to become dependent on state benefits. In addition,
financial education initiatives might help reducing the dispersion in wealth; a dispersion that
is much higher than the often debated inequality in income (Cagetti and De Nardi, 2006).
This paper is organized as follows. In Section 2, we review the current literature on
wealth accumulation in relation to financial sophistication. In Section 3, we present data and
descriptive statistics, and explain how the measures of financial ability and sophistication are
constructed. In Section 4, we report the results of wealth regressions including measures of
financial ability and sophistication. In Section 5, we present several extensions and discuss
the robustness of the results. In Section 6, we consider two channels by which financial skills
96
exert an effect on wealth accumulation: stock market participation and retirement planning
activities. In addition, we examine the economic relevance of being financially sophisticated.
In Section 7, we conclude with some remarks on implications for policy and areas for future
research.
2. Literature
The simplest version of the life cycle consumption model without bequests and
uncertainty predicts that households accumulate savings during their working career to
finance retirement and decumulate wealth thereafter (Modigliani and Brumberg, 1954). This
type of savings behavior enables households to smooth their marginal utility of consumption
over the life cycle. However, there are many reasons why household consumption and wealth
follow different patterns and the standard model can quite easily be adjusted to cope with
many of them (Browning and Lusardi, 1996; Cagetti and De Nardi, 2006).
A large variety of empirical research sheds light on the observed patterns in wealth
dispersion and portfolio choice. Studies have highlighted among others the role of
precautionary savings motives (Hubbard, Skinner and Zeldes, 1995), longevity and bequests
(Hurd, 1989), different economic opportunities across cohorts (Kapteyn, Alessie and Lusardi,
2005), self-control (Laibson, 1997; Benartzi and Thaler, 2004; Ameriks, Caplin, Leahy and
Tyler, 2007), unexpected events (Venti and Wise, 2000; Lusardi, 2003), background income
risk (Heaton and Lucas, 2000; Guiso, Jappelli, Terlizzese, 1996), and health (Rosen and Wu,
2004). To the best of our knowledge, none of these studies focus on the role of financial
capabilities in accumulating savings, while more financially sophisticated individuals are
likely to perceive lower barriers for gathering and processing information and are thus better
equipped to manage their savings portfolio. Somewhat related to the subject of our study is
the work by Chan and Stevens (2008) who document that households base pension and
retirement savings decisions upon the limited and sometimes incorrect pension knowledge
they have.2
Bernheim (1995, 1998) was among the first to stress that policymakers and
researchers might have overlooked the importance of financial literacy for savings. Since then
many studies emphasize the role of financial sophistication but, in absence of specific literacy
measures, resort to crude proxies for financial skills, such as income, wealth or education
2 Many authors have documented that households are rather ill-informed about their Social Security benefits and company pensions. See Gustman and Steinmeier (2004) and Van Els, Van den End and Van Rooij (2004) for evidence for the US and the Netherlands, respectively.
97
(Calvet, Campbell and Sodini, 2007; Vissing-Jorgenson, 2004). The disadvantage of these
proxies is that there is no way to disentangle the effect of financial ability from the effect of
the proxy variable. By using education as a measure of financial sophistication one is not able
to separate the independent effect of financial skills from the impact of the education level as
such, which in many regression specifications also serves as a proxy for lifetime income.
Christiansen, Joensen and Rangvid (2008) use information on formal economics education as
an alternative proxy for individual financial sophistication to study portfolio decisions.
Since a few years researchers have increased effort in developing specific measures of
financial ability and knowledge and have started investigating its relation to economic
decisions and portfolio choice. Hilgert, Hogarth and Beverly (2003) explore the relation
between literacy and money management, while Lusardi and Mitchell (2006) consider the
associations with retirement planning. More recently Van Rooij, Lusardi and Alessie (2007)
and Christelis, Jappelli and Padula (2007) have studied the link between the decision to invest
in stocks and specific measures of financial sophistication and basic cognitive ability.
Several authors have stressed that the welfare costs of financial mistakes are not
negligible (Campbell, 2006; Calvet, Campbell and Sodini, 2007; Cocco, Gomes and
Maenhout, 2005). Nevertheless, an increasing amount of studies documents the prevalence of
financial mistakes. Agarwal, Driscoll, Gabaix and Laibson (2007) provide evidence of
financial mistakes in the loan market with many households paying too much fees or too high
interest rates on credit card debt, home equity loans and mortgages (see also Moore, 2003).
Calvet, Campbell and Sodini (2007) show that in Sweden – a country that is claimed to have
efficient investors – many households hold underdiversified portfolios or do not participate in
financial markets at all.
The amount of financial mistakes might not come as a surprise given the body of
evidence on limited financial literacy among households. This evidence is robust in different
settings and across different countries of which many have reacted by setting up financial
education programs (OECD, 2005). While the large variation in the initiatives to enhance
awareness and financial sophistication creates new possibilities to learn how to effectively
design and implement education programs in the near future, these evaluations have been
limited so far (Smith and Stewart, 2008).
The impact of financial education on savings behavior has been investigated almost
exclusively in the context of retirement seminars offered by US firms. An important exception
is the work by Bernheim, Garrett and Maki (2001) who document positive effects of financial
education during high school on long term savings employing the variability in state mandates
98
on the teaching of topics related to household financial decisions. Bernheim and Garrett
(2003), Lusardi (2004) and Clark and D’Ambrosio (2008) have documented positive effects
of retirement seminars in the workplace, especially when it regards the intentions to change
savings behavior. Overall, however, the evidence is mixed as other studies were not able to
come up with significant, lasting effects (Duflo and Saez, 2003, 2004).
Moreover, as the attendance in retirement seminar is voluntary it is not to be excluded
that participants form a selected group that is already more intrinsically motivated to remedy
insufficient pension savings. In addition, any beneficial effect of retirement seminars could
also be the direct result of the provision of information on the need for retirement savings
rather than of an increase in financial sophistication. This is especially likely as retirement
seminars typically take a few hours at most. Interestingly, Mandell (2008) does not find a
literacy enhancing effect of more intensive courses at high school devoted to teaching
personal finance and money management on test scores for financial literacy. This suggests
that the effect of financial education in high school or at retirement seminars on savings could
also work via other channels than raising financial knowledge and ability. The impact of
financial education on savings in these studies might for example work more indirectly
through an effect on individual characteristics and the appetite for saving. In this paper, we do
not evaluate financial education programs but focus directly on the role of actual financial
knowledge and capabilities in wealth accumulation and disentangle its effects from other
personal traits including risk tolerance, patience, and other preferences related to the
propensity to save.
3. Data
We have devised a special module for the annual DNB Household Survey (DHS)
including an elaborate set of questions on financial ability and knowledge as well as a section
on retirement planning activities. The questions have been answered by the household panel
run by CentERdata; a survey agency at Tilburg University specialized in internet surveys.3 It
is important to note that - even though the Netherlands has an internet penetration of about
80% - the selection of panel members is not dependent on the use and availability of internet.
Households without a computer or an internet connection are provided with the necessary
equipment (e.g. a set-top box to participate through their television connection). Attrition is
dealt with by biannual refreshment samples that are drawn in view of keeping the panel
3 For more information, we refer to http://www.uvt.nl/centerdata/en.
99
representative of the Dutch population of 16 years and older (persons staying in hospitals,
specialized care institutions or prisons are not included).4
The questionnaire was held among those persons within the household who are in
charge of household finances. It was fielded in 2005 from September 23 until September 27
and repeated a week thereafter for those households that had not responded yet. The response
rate equaled 74.4% (1508 out of 2028 households). The DHS contains a lot of information on
income and work, health, household debt and assets, and an extensive set of psychological
questions on attitudes with respect to saving and portfolio investments.5 We merge our
module on financial literacy with the data in the 2005 wave of DHS on net worth for those
households for who we have information on all of their assets and debts. Since wealth
regressions might be sensitive to outliers we trim the net worth variable by excluding the top
and bottom 1% of observations which are most suspicious to measurement error.
After these steps, our reduced sample consists of 1091 households. The average age of
the respondents equals 50.8 (ranging from 22 to 90 years); 53.1% of the respondents are male;
56.7% are married or living together with a partner, about one third have children living at
home and 20.4% of the respondents are retired. Comparison of these characteristics with the
full sample shows that especially elderly respondents report their asset and debt position more
frequently, but overall the composition of the sample remains fairly unchanged. Table 1
reports the median, mean and standard deviation of household net worth. This wealth concept
includes all kind of private savings and investments accounts, housing wealth, other real
estate, and durable goods, net of mortgages and other financial debt. It is clear that its
distribution is skewed and that there is a lot of dispersion in net worth also after the
substantial reduction due to the trimming process.
This paper aims at exploring a new potential explanation contributing to the
heterogeneity in wealth holdings, i.e. the role of the apparent widespread differences in
financial literacy. First, we look at the bivariate relationship between wealth holdings and two
financial literacy indices which have been derived from our financial literacy module (Table
2). The basic literacy index is a measure for very basic financial ability and knowledge and
follows from a factor analysis based on the correct answers to five simple questions on the
understanding of inflation, interest rates and interest compounding. The advanced literacy
index is based on a factor analysis using the information content of correct, incorrect and do
4 In addition, we use household weights to calculate descriptive statistics to ensure representativeness of the population. 5 Direct information on consumption and annual saving out of income is not available.
100
not know answers to eleven questions on financial knowledge about the understanding of
stocks, bonds and mutual funds, their trade off between risk and return and the benefits of risk
diversification. The exact wording of these questions, the response patterns, an explanation of
the construction of these indices and the relation to demographics is documented in detail in a
previous paper (Van Rooij, Lusardi and Alessie, 2007) which is also included as the second
paper in this thesis.
Table 1. Total net worth statistics Thousands of euro ________________________________________________________________ Total net worth ______________________________________ Total net worth Median Mean Standard deviation _______________________ _________ _________ _______________ before trimming (N=1116) 119.7 184.3 279.3 after trimming (N=1091) 119.7 167.1 189.0 ________________________________________________________________ Table 2. Total net worth versus basic and advanced literacy Thousands of euro (N=1091) ________________________________________________________________ Total net worth ______________________________________ Basic literacy quartiles Median Mean Standard deviation _______________________ _________ _________ _______________ 1 (low) 43.9 117.2 162.3 2 98.8 150.2 164.7 3 111.2 156.5 173.6 4 (high) 142.8 195.7 209.3 Total net worth ______________________________________ Advanced literacy quartiles Median Mean Standard deviation _______________________ _________ _________ _______________ 1 (low) 46.7 100.1 121.2 2 82.0 129.3 151.0 3 112.4 167.5 181.4 4 (high) 185.9 236.3 228.4 ________________________________________________________________
Table 2 documents a strong increase in median net worth with basic and advanced
literacy. The median net worth position of the top quartile of financially sophisticated
individuals amounts to €185900 which is the quadruple of the median net worth position in
the bottom advanced literacy quartile (€46700). Also the differences in wealth position across
101
basic literacy quartiles are large - although somewhat smaller than for advanced literacy.
These simple correlations suggest a strong, non-linear gradient between literacy and net
worth.
Table 3 shows a similar pattern for several asset categories. Home ownership and
investments in stocks, mutual funds and bonds are much more common among those who
score high on the literacy scales. Nevertheless there are obvious differences between asset
classes. While home ownership is also not uncommon among the most illiterate households,
investments in individual stocks or bonds are almost absent within this subgroup. This
evidence suggests that more literate households hold more diversified portfolios or at least
spread their wealth over a richer class of assets.
Table 3. Asset ownership versus basic and advanced literacy Weighted percentages (N=1116) ____________________________________________________________________________ % of households owning __________________________________________________ Basic literacy quartiles Stocks Mutual funds Bonds Home _______________________ ___________ ___________ ___________ ___________ 1 (low) 2.4 5.6 1.9 40.5 2 9.7 17.6 3.8 53.4 3 10.2 16.5 3.0 54.4 4 (high) 18.1 23.9 6.1 60.8 % of households owning __________________________________________________ Advanced literacy quartiles Stocks Mutual funds Bonds Home _______________________ ___________ ___________ ___________ ___________ 1 (low) 2.0 6.5 1.4 44.6 2 5.0 11.8 1.2 44.8 3 14.2 18.5 5.0 56.0 4 (high) 25.2 33.1 8.8 70.9 ____________________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding. 4. Wealth regressions
To further investigate the relation between wealth accumulation and financial
sophistication, we start with a basic multivariate regression for total net worth and extend this
specification by successively including additional information. Tables 4A and 4B report the
results. First, we run an OLS regression of total net worth on our measure for basic financial
skills and cognitive ability. Other control variables include gender, age and education level of
102
the respondent, household composition (marital status and the number of children within the
household), household net disposable income, and a dummy for whether the respondent is
retired. We have also included a dummy for being self-employed as entrepreneurs differ in
many aspects from others and might behave accordingly (Hurst and Lusardi, 2004).
Table 4A. Total net worth and financial literacy: multivariate regressions __________________________________________________________________________________________ (1) (2) (3) OLS OLS OLS __________________ __________________ ___________________ Basic financial literacy index 12328*** (3.42) 15804*** (3.37) 15712*** (3.08) Age dummy (30<age<=40) 26904** (2.25) 24581** (2.02) 22398* (1.69) Age dummy (40<age<=50) 72269*** (5.42) 72359*** (5.34) 74986*** (5.20) Age dummy (50<age<=60) 131181*** (8.71) 130456*** (8.49) 136511*** (8.33) Age dummy (60<age<=70) 143929*** (7.01) 144246*** (6.94) 152902*** (7.25) Age dummy (age>70) 166320*** (6.31) 161898*** (5.88) 168605*** (6.15) Intermediate vocational education 18230 (1.37) 12666 (0.93) 12961 (0.92) Secondary pre-university education 10709 (0.65) 2851 (0.18) 4714 (0.28) Higher vocational education 25853* (1.85) 22434 (1.59) 18835 (1.30) University education 37059** (1.98) 35853* (1.88) 26112 (1.32) Male -7952 (0.81) -10204 (1.02) -20710** (1.97) Married 30905*** (2.72) 26639** (2.29) 24494** (2.08) Number of children 10285* (1.70) 11166* (1.80) 10199 (1.59) Retired 45437** (2.16) 45454** (2.11) 42855** (2.03) Self-employed 26205 (1.17) 25016 (1.12) 25300 (1.04) Ln(household income) -3277982*** (3.76) -3261105*** (3.72) -3062710*** (3.69) Ln2(household income) 315864*** (3.71) 314721*** (3.67) 297871*** (3.67) Ln3(household income) -9676*** (3.51) -9648*** (3.45) -9179*** (3.48) High confidence in financial skills -10738 (0.79) -9253 (0.66) Low confidence in financial skills -26368** (2.15) -21614* (1.70) Risk aversion dummy 2 (low) -1181 (0.043) Risk aversion dummy 3 -16204 (0.65) Risk aversion dummy 4 -30789 (1.24) Risk aversion dummy 5 -13917 (0.53) Risk aversion dummy 6 -55402** (2.41) Risk aversion dummy 7 (very high) -64013*** (2.85) Constant 10880396*** (3.67) 10818615*** (3.65) 10088240*** (3.58) Observations 1091 1060 1013 R-squared 0.32 0.32 0.34 p-value test age=0 0.00 0.00 0.00 p-value test education=0 0.26 0.27 0.61 p-value test income=0 0.00 0.00 0.00 p-value test confidence=0 0.10 0.24 p-value test risk aversion=0 0.00 __________________________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is net worth in thousands of euro. The most risk tolerant, none smoking and moderately drinking (4 alcoholic drinks or less a day) respondents are in the reference group.
Age and income appear to be strongly significant (Table 4A, column 1). Total net
worth is increasing in age, but using cross-section data we cannot disentangle whether this is
103
attributable to age or cohort effects. Nevertheless, it is consistent with panel data evidence
suggesting that Dutch households hardly decumulate private wealth after retirement (Kapteyn,
Alessie and Lusardi, 2005; Alessie, Lusardi and Kapteyn, 1999).6 To capture complex,
possibly non-linear effects of income on wealth accumulation, we have included a polynomial
for the natural logarithm of net disposable household income with a linear, quadratic and
cubic term. A one percent increase in household income – measured at mean levels of the
control variables – is associated with an increase in total net worth by somewhat more than
€1400.
Most interesting is the positive and significant effect of basic cognitive financial
ability on total net worth. A unit increase in basic literacy goes together with about €12000
more wealth (the basic literacy measure itself has a zero mean and a standard deviation of
one). Individuals with higher cognitive ability seem to be more likely to accumulate savings.
Nevertheless, it is not immediately clear whether this is the result of better financial decisions
because of the ability to collect and process information at low cost and effort or runs through
its association to personal characteristics like risk aversion, time preference or overconfidence
(see for example Christelis, Jappelli and Padula (2007) for a discussion).
To further investigate these issues, we first examine the role of confidence in financial
skills in relation to actual financial knowledge. In addition to actual financial ability,
perceptions of one’s own ability might assert an independent effect on financial outcomes
albeit the direction of the effect is not clear-cut a priori. Persons who are overly modest about
their skills might refrain from financial innovations and forego potential financial benefits.
Insofar high confidence in one’s personal skills leads to less conservative portfolio
management it could have a positive impact on net worth. On the other hand, these people
might buy complex products that they do not fully understand and could end up making
financial mistakes with serious money at stake. In addition, in the literature on overconfidence
it is argued that individuals with too much trust in their own skills could be inclined to
interpret and filter information in accordance with their beliefs and might trade excessively
(ending up with high trading costs and lower net investment returns). Barber and Odean
(2000, 2001) for instance provide evidence of overconfident investors trading excessively and
ending up with lower returns.
At the start of our survey, we ask respondents ‘How would you assess your
understanding of economics (on a 7-points scale; 1 means very low and 7 means very high)?’
6 The increase in the 70 plus age group could also be partly related to different mortality rates depending upon the wealth position (Hurd, 1990).
104
Based upon this self-assessment of economic literacy, we construct a relative measure of
overconfidence. The self-reported literacy question and our basic ability index are not directly
comparable due to the use of different scales, but do provide information on the relative
position of respondents within the distribution of actual basic literacy and self-assessed
literacy, respectively. We start with grouping both variables into four categories and rank the
respondents accordingly from the top category to the lowest group. Thereafter, we create a
dummy for overconfidence that equals unity if the respondents’ self-assessed literacy ranking
is higher than our classification for basic financial skills. In addition, we construct a dummy
for relatively low confidence or underconfidence measuring whether the ranking on self-
assessed literacy is more modest than warranted. Thereafter, we rerun the first wealth
regression now including both dummies (the reference group being the respondents with a
proper assessment of their skills). Append A provides more details on the construction of the
confidence measures. Our main interest is whether the effect of basic financial ability on
wealth accumulation is affected by the inclusion of the confidence measures. The coefficient
of basic financial capabilities remains significant and increases somewhat (Table 4A, column
2).7 The coefficient of overconfidence is negative but insignificant. Underconfidence however
has a significant negative impact on net worth. Compared to persons with proper knowledge
of their financial skills, these people do not seem to take full advantage of their capabilities.
Experimental evidence reveals that individuals with lower cognitive ability are likely
to be less risk tolerant and more impatient (Benjamin, Brown and Shapiro, 2006; Dohmen,
Falk, Huffman and Sunde, 2007). To test whether the effect of cognitive ability runs through
an association with risk attitude, we include a measure of risk aversion. In the annual DHS
respondents are asked to indicate to what extent they agree with the following statement
‘Investing in stocks is something I don’t do, since it is too risky’. The response scale runs from
1 to 7, where 1 corresponds to complete disagreement and 7 to complete agreement. Kapteyn
and Teppa (2002) use this measure and show that it has more explanatory power in models of
portfolio choice than measures of risk tolerance based on a series of hypothetical choices
between uncertain streams of lifetime income as introduced by Barsky, Juster, Kimball and
Shapiro (1997). The regression results in Table 4A (column 3)8 show that there is indeed an
7 The number of observations has now decreased from 1091 to 1060 as, in constructing the measures for under and overconfidence, we ignore respondents answering ‘do not know’ when asked to assess their financial skills. 8 The information on risk aversion and time preferences is available in the DHS modules on savings attitudes, income and health. Due to the merging process the total number of observations in our regression is reduced by 57 (even though we were able to retain some households by using information on time preferences and risk tolerance from adjacent years).
105
important role for risk aversion in explaining wealth heterogeneity, but the coefficient of basic
financial skills is virtually unaffected.9
We subsequently test whether financial ability serves as a proxy for patience. We do
not have direct information on time preferences, but we include information on smoking and
drinking behavior as a proxy for myopic behavior as it is done in many other studies since the
work by Fuchs (1980) on the relation between different types of health decisions and
patience. We use information on whether people smoke and how often, and on whether they
are heavy drinkers (more than four alcoholic drinks on average per day). We do not find any
relation between net worth and these proxies for time preference and the coefficient of the
basic financial literacy index changes only marginally (Table 4B, column 1)
In the next step, we investigate whether basic financial ability could be a proxy for
more advanced financial skills (as suggested by the results in Van Rooij, Lusardi and Alessie,
2007) and include the measure of advanced financial sophistication. Indeed the effect of
advanced literacy is strongly significant, reduces the coefficient on basic financial capacity
and wipes out its significance (Table 4B, column 2). The coefficient of advanced literacy is
higher than the original effect of the basic ability index and a unit increase in financial
sophistication raises the household net worth position by €24000 on average. However, we
need to be cautious with the interpretation of the OLS coefficient for financial sophistication.
While the financial ability index touches upon very basic cognitive skills that people more or
less need on a daily basis, the advanced literacy index includes questions on the working of
stocks, bonds and mutual funds and addresses skills which are not a necessity in daily
transactions. It is conceivable that wealth management fosters the collection of financial
knowledge and the OLS coefficient could be biased upwards (simultaneity bias). On the other
advanced literacy index might be a noisy measure of actual financial skills and the coefficient
of financial sophistication could be biased to zero (attenuation bias). Indeed Van Rooij,
Lusardi and Alessie (2007) provide evidence of the importance of slight variations in the
wording of questions for response patterns, which suggests that there is some guessing going
on for questions that appear hard to grasp.
9 As a robustness check we have included the Barsky et al. (1997) measure of risk tolerance as it has proved to be a valuable measure in other papers (e.g. Van Rooij, Prast and Kool, 2007), but it turned out to be insignificant confirming the results of Kapteyn and Teppa (2002).
106
Table 4B. Total net worth and financial literacy: multivariate regressions __________________________________________________________________________________________ (1) (2) (3) OLS OLS IV __________________ __________________ ___________________ Advanced financial literacy index 23514*** (4.86) 67122** (2.28) Basic financial literacy index 16694*** (3.17) 9050 (1.64) -5129 (0.45) Age dummy (30<age<=40) 20743 (1.55) 24756* (1.81) 32198** (2.12) Age dummy (40<age<=50) 76027*** (5.24) 77806*** (5.31) 81106*** (5.24) Age dummy (50<age<=60) 136072*** (8.17) 134470*** (8.05) 131499*** (7.49) Age dummy (60<age<=70) 151976*** (7.18) 150595*** (7.11) 148034*** (6.71) Age dummy (age>70) 169144*** (6.16) 169701*** (6.17) 170733*** (6.08) Intermediate vocational education 16282 (1.14) 12459 (0.87) 5368 (0.35) Secondary pre-university education 5994 (0.35) -1197 (0.07) -14533 (0.76) Higher vocational education 17733 (1.21) 11324 (0.77) -563 (0.03) University education 25821 (1.30) 16848 (0.84) 208 (0.01) Male -19907* (1.84) -26884** (2.49) -39823*** (3.01) Married 22754* (1.89) 24778** (2.07) 28533** (2.28) Number of children 10687* (1.66) 11424* (1.79) 12790** (1.99) Retired 43503** (2.06) 41651** (1.98) 38215* (1.78) Self-employed 26025 (1.07) 24797 (1.03) 22520 (0.93) Ln(household income) -3066220*** (3.68) -3011077*** (3.57) -2908803*** (3.28) Ln2(household income) 299340*** (3.66) 293782*** (3.57) 283474*** (3.30) Ln3(household income) -9261*** (3.48) -9084*** (3.40) -8754*** (3.17) High confidence in financial skills -8685 (0.61) -9829 (0.70) -11951 (0.84) Low confidence in financial skills -23286* (1.83) -19605 (1.55) -12778 (0.94) Risk aversion dummy 2 (low) -3888 (0.14) -8001 (0.29) -15629 (0.57) Risk aversion dummy 3 -21340 (0.86) -23968 (0.97) -28841 (1.17) Risk aversion dummy 4 -35329 (1.41) -33869 (1.36) -31162 (1.23) Risk aversion dummy 5 -16025 (0.60) -19345 (0.74) -25502 (0.99) Risk aversion dummy 6 -57751** (2.51) -54037** (2.37) -47149** (1.98) Risk aversion dummy 7 (very high) -66105*** (2.93) -60545*** (2.71) -50234** (2.07) Smoking: every now and then -20230 (1.22) -18589 (1.15) -15544 (0.95) Smoking: daily (< 20 cigarettes) -6861 (0.39) -5978 (0.34) -4339 (0.25) Smoking: daily (>= 20 cigarettes) -20227 (0.73) -21097 (0.76) -22711 (0.82) Drinking: daily (> 4 drinks) -966 (0.04) -1802 (0.08) -3353 (0.15) Constant 10066777*** (3.56) 9897789*** (3.45) 9584366*** (3.15) Observations 1003 1003 1003 R-squared 0.34 0.35 0.32 p-value test age=0 0.00 0.00 0.00 p-value test education=0 0.64 0.81 0.84 p-value test income=0 0.00 0.00 0.00 p-value test confidence=0 0.18 0.30 0.56 p-value test risk aversion=0 0.00 0.01 0.48 p-value test smoking, drinking=0 0.74 0.77 0.83 p-value Hansen J test 0.30 F-statistic first stage regression 13.0 p-value exogeneity test 0.18 __________________________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is net worth in thousands of euro. The most risk tolerant, none smoking and moderately drinking (4 alcoholic drinks or less a day) respondents are in the reference group. The advanced literacy index has been instrumented using dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
Therefore, we perform an instrumental variables (IV) regression including economics
education as an instrument for advanced financial literacy. This variable measures the
107
exposure to education before entering the job market. It is based upon the answers to the
question ‘How much of your education was devoted to economics?’ where response
categories include the options ‘a lot’, ‘some’, ‘little’, and ‘hardly at all’.10 We assume that this
information is unrelated to wealth. It has strong predictive power for financial literacy as
shown by the test on instrument relevance in the first stage regression (Table 4B, column 3).
The F-value equals 13, clearly above 10 the value that is often recommended as a rule of
thumb to be sure that problems due to weak instruments are avoided (Staiger and Stock,
1997). The estimation results show that the IV coefficient remains significant at the 5% level
and increases substantially to €67000 suggesting that financial literacy is indeed measured
with imprecision. The Hansen J-test on the validity of the overidentifying restrictions is not
rejected. Overall, our estimates are in line with the hypothesis that financial sophistication is
an important determinant of wealth accumulation also after accounting for attitudes and
preferences which might be associated with the level of financial sophistication.
5. Extensions
One potential concern with our instrument is that accumulating wealth, and becoming
literate or being exposed to economics education are choice variables depending on a
common unobserved factor or another omitted variable. One possible candidate for a variable
that drives literacy, education and wealth but is usually unavailable in wealth regressions is
ability as some people are intrinsically more gifted by nature with talent and basic cognitive
skills then others. For this reason precisely we maintain the basic literacy variable in the
wealth regressions as to control for cognitive ability.
Carefulness is an example of an important common factor that is perhaps not
sufficiently taken into account yet. Careful persons taking many precautions to prevent bad
thing happening to them could be more likely to hold additional savings buffers and to invest
in financial education as well to lower the chance to enter a debt situation or end up with
financial problems. To explore this possibility we run two additional specifications including
information from two separate questions on whether respondents consider themselves a
‘careful person’, and whether they ‘take many precautions’. The response scales run from 1
(completely disagree) to 7 (completely agree). Appendix B reports the precise wording of
these questions, which are available in a separate DHS module. By merging this information
with our data we lose close to 300 observations. Due to the lower number of observations, the
F-value of the joint significance of economics education in the first stage regression decreases 10 See appendix B for the precise wording.
108
to 6 but remains strongly significant. More importantly, Table 5A shows that the inclusion of
how careful the respondents are does not take away anything from the effect of financial
sophistication on net worth. The advanced literacy coefficient remains significant at the 5%
confidence level and even increases in value.
Table 5A. Total net worth regressions: including carefulness and precaution _____________________________________________________________________________
(1) (2) (3) (4) OLS IV OLS IV __________ __________ __________ _________ Advanced financial literacy index 24139*** 92061** 25390*** 96858** (4.03) (2.19) (4.26) (2.33) Basic financial literacy index 10023 -12794 10813* -13227 (1.60) (0.82) (1.68) (0.85) Carefulness dummy 2 (low) -43822 -40941 (1.18) (1.13) Carefulness dummy 3 -50935 -33725 (1.48) (0.97) Carefulness dummy 4 -10235 3741 (0.30) (0.11) Carefulness dummy 5 6059 10025 (0.17) (0.29) Carefulness dummy 6 (very high) -6969 -8211 (0.19) (0.23) Precaution dummy 2 (low) 24382 1035 (0.64) (0.024) Precaution dummy 3 7903 5677 (0.24) (0.16) Precaution dummy 4 25802 16869 (0.80) (0.48) Precaution dummy 5 19022 5463 (0.59) (0.15) Precaution dummy 6 (very high) 29969 29647 (0.88) (0.82) Demographics (see table 4B) yes yes yes yes Observations 721 721 721 721 R-squared 0.38 0.31 0.37 0.29 p-value test carefulness=0 0.00 0.03 p-value test precaution=0 0.80 0.78 p-value Hansen J test 0.14 0.15 F-statistic first stage regression 6.24 6.12 p-value exogeneity test 0.12 0.10 ______________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is net worth in thousands of euro. The reference group in the first two columns contains those respondents who strongly disagree with the statement that they consider themselves as a careful person. The reference group in the last two columns contains those respondents who strongly disagree with the statement that they take many precautions. Other control variables, not reported for brevity, are the same as in Table 4B. The advanced literacy index has been instrumented using dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
109
Another potential concern with respect to our result that financial sophistication leads
to higher net wealth holdings is that net worth is a very heterogeneous concept. Although we
have included controls for the impact of demographics, risk aversion, time preferences and
confidence measures many other potential drivers of wealth heterogeneity could be related to
financial sophistication - possibly in an unexpected way - and might influence the relation
between financial literacy and the accumulation of savings. In this section we further exploit
the richness of the DHS dataset to investigate whether the importance of the effect of financial
sophistication is taken away once we control for alternative explanations of wealth dispersion.
One potential explanation for wealth heterogeneity is simply that households have a
different appetite for saving. Venti and Wise (1998, 2000) for example conclude that
unobserved heterogeneity in the taste for saving must be a major driving factor for wealth
inequality after having eliminated successively lifetime earnings, chance events and
investment choices as sufficient explanations. Our dataset does contain a direct proxy for the
appetite for saving; we include the responses to the question on what respondents ‘do with
money that is left over after having paid for food, rent, and other necessities’. The response
scale runs from 1 to 7, where 1 means ‘I like to spend all my money immediately’ and 7 means
‘ I want to save as much as possible’. Exact wording and responses are reported in appendix
B. Table 5B (columns 1 and 2) indeed shows that across the board a higher taste for saving
translates into more accumulated savings. Being a crude proxy that perhaps could also serve
as a measure of patience, the most important result from the table is that the magnitude and
significance of the coefficient of financial sophistication is unaffected.
Another alternative measure for time preference can be obtained from the question
whether people use a short or a long forward looking horizon in their spending decisions.
Being a direct measure of patience and saving compared to the commonly used smoking and
drinking proxies for time preference, the disadvantage is that responses to this question could
be related to a number of other personal characteristics and background information. That said
the estimates show that the responses have clear predictive value for wealth accumulation
(Table 5B, columns 3 and 4). Nevertheless, the inclusion of this measure does not take away
the effect of financial sophistication on net worth.
Self-control is indisputably an important factor in savings outcomes (Thaler, 1994).
No matter how much importance people attach to savings, if they have difficulties to
withstand the short term temptations of consumption and cannot find ways to constrain their
consumption behavior, they will hold savings below their target level. The question to
respondents whether they ‘find it difficult to control their expenditures’ (on a scale from 1 to 7
110
where 1 means ‘very easy’ and 7 means ‘very difficult’) appeals directly to problems of self-
control. As expected self-control is a major determinant for wealth accumulation (Table 5C,
columns 1 and 2). The difference in net worth between those who have little or no problems
in controlling their expenses and those who recognize that this is a major challenge is as much
as nearly €90000. The inclusion of self-control, however, does not fundamentally affect the
relation between financial literacy and wealth accumulation.
Table 5B. Total net worth: including taste for saving and alternative time preferences
_____________________________________________________________________________ (1) (2) (3) (4) OLS IV OLS IV __________ __________ __________ __________ Advanced financial literacy index 20951*** 63127** 23189*** 64954** (4.40) (2.15) (4.85) (2.21) Basic financial literacy index 6763 -6445 10022* -3589 (1.21) (0.58) (1.81) (0.31) Taste for saving dummy 2 (low) 41138** 31847 (2.20) (1.61) Taste for saving dummy 3 52947*** 47649*** (3.18) (2.76) Taste for saving dummy 4 68209*** 61623*** (4.37) (3.79) Taste for saving dummy 5 100078*** 86603*** (5.94) (4.53) Taste for saving dummy 6 (very high) 68491*** 57392*** (3.42) (2.62) Time preference: horizon 3-12 months 663 -939 (0.053) (0.074) Time preference: horizon 1-5 years 32813** 33408*** (2.56) (2.59) Time preference: horizon 5-10 years 55025*** 52812*** (2.67) (2.59) Time preference: horizon > 10 years 60375** 55111** (2.32) (2.08) Demographics (see table 4B) yes yes yes yes Observations 1003 1003 1003 1003 R-squared 0.37 0.34 0.37 0.34 p-value taste for saving=0 0.00 0.00 p-value test time preference=0 0.00 0.00 p-value Hansen J test 0.33 0.26 F-statistic first stage regression 12.6 13.0 p-value exogeneity test 0.20 0.22
______________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is net worth in thousands of euro. The reference group in the first two columns contains those respondents with a very low taste for saving. The reference group in the last two columns contains those respondents with a very sort time horizon (a couple of months). Other control variables, not reported for brevity, are the same as in Table 4B. The advanced literacy index has been instrumented using dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
111
Table 5C. Total net worth regressions: including self-control and bequest motives _____________________________________________________________________________
(1) (2) (3) (4) OLS IV OLS IV __________ __________ __________ __________ Advanced financial literacy index 21539*** 63363** 18918*** 71014** (4.47) (2.18) (4.04) (2.45) Basic financial literacy index 5950 -7420 8797 -7154 (1.06) (0.66) (1.63) (0.66) Self-control dummy 2 (quite easy) -13081 -9695 (0.79) (0.58) Self-control dummy 3 -43830** -35643* (2.45) (1.84) Self-control dummy 4 -47582** -39237* (2.46) (1.95) Self-control dummy 5 -68355*** -58363*** (3.86) (2.99) Self-control dummy 6 (quite difficult) -88070*** -86862*** (4.48) (4.41) Dummy bequest motive: yes 106732*** 103244*** (4.81) (4.66) Dummy bequest motive: no -12838 -10935 (0.88) (0.73) Dummy bequest motive: other -57490*** -32600 (2.87) (1.26) Demographics (see table 4B) yes yes yes yes Observations 1003 1003 1003 1003 R-squared 0.37 0.34 0.40 0.36 p-value taste self-control=0 0.00 0.00 p-value test bequest motive=0 0.00 0.00 p-value Hansen J test 0.21 0.29 F-statistic first stage regression 13.4 12.8 p-value exogeneity test 0.23 0.08
______________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is net worth in thousands of euro. The reference group in the first two columns contains those respondents who find it very easy to control their expenditures. The reference group in the last two columns contains those respondents who do not have children. Other control variables, not reported for brevity, are the same as in Table 4B. The advanced literacy index has been instrumented using dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
The same is true if we take into account that bequest motives might be associated with
vast differences in wealth accumulations. Although there is no a priori reason to believe that
financial sophistication is related to the intention to leave bequests, the bequest motive might
be an omitted variable explaining a large part of the variation in wealth accumulation. Indeed
the empirical results suggest that some households hold substantial amounts of their wealth
for intentional bequests (Table 5C, columns 3 and 4). The positive impact of financial
sophistication on net worth survives upon inclusion of the bequest motive: its magnitude and
significance even increase somewhat.
112
In addition to these extensions we have incorporated a large number of variables
which based upon the theoretical and empirical literature could principally account for part of
the variation in net worth among households. To this end, we have utilized the rich dataset we
have available by merging our survey data with information from other DHS-modules which
inevitably sometimes leads to a loss of observations. At the same time the variables employed
are sometimes simple, crude proxies but may serve at least as a first test for the underlying
hypotheses. We have included several alternative health measures, the self-assessed
probability of the respondent for survival until certain age levels to account for heterogeneity
with respect to perceived longevity, income uncertainty, expectations regarding house price
developments, the perceived likelihood of a future reduction in the generosity of the state
pension, and the expected replacement rate (based upon state pension eligibility and
mandatory employer company savings). The latter proxies annuitized pension wealth which is
not part of the private net worth position. All these variables appear insignificant and do not
take away the effect of financial sophistication. Finally, we have tested the robustness of our
results to other specifications of the wealth regression. Using net worth over permanent
income as a dependent variable, where permanent income is calculated from an auxiliary
regression of income on a number of demographics, gives estimation results which
corroborate the evidence of a positive and significant impact of financial sophistication on
wealth.
6. Discussion
6.1 Financial sophistication and stock market participation
Given that financial sophistication increases household wealth holdings, it might be
attractive from a public policy point of view to invest in financial education initiatives. To
learn about what type of education programs might be most successful it is important to
understand the mechanisms at work behind the relation between financial sophistication and
net worth. We explore two possible explanations related to the well documented limited stock
market participation puzzle and to another puzzling fact in household finance, i.e. the lack of
retirement planning.
Economic theory dictates that possibly except for a small proportion of households it
is optimal to hold at least part of their wealth in the form of stocks (Haliassos and Bertaut,
1995). Investments in the stock market provide the opportunity to exploit the equity premium
and to benefit from risk diversification. International evidence on the composition of
113
household portfolios shows that many households have no stocks at all in their wealth
portfolio (Guiso, Haliassos and Jappelli, 2002). In our sample about a quarter of the
households invest in stocks either direct or indirect via mutual funds. The limited participation
in stock markets is mostly explained by transaction costs and the costs of processing
information which create a threshold for entering the stock market (Haliassos and Bertaut,
1995; Vissing-Jorgenson, 2004). In addition, it has been argued that households are either
simply unaware of the opportunities to invest in stock markets or refrain from doing so due to
a lack of trust (Guiso and Jappelli, 2005; Guiso, Sapienza and Zingales, 2008).
An increase in financial sophistication lowers information costs as well as
impediments to participating due to a lack of knowledge or trust in the working of financial
markets. Indeed, the regression results reported in Van Rooij, Lusardi and Alessie (see the
second paper of this thesis) show that the probability to own stocks or mutual funds increases
by about 8 percentage points upon a one-standard deviation increase in the level of financial
sophistication, and about 16 percentage points when we use economics education as an
instrument for financial literacy. The latter corrects for measurement error in the index for
advanced financial literacy and takes into account that one might accumulate financial
knowledge in the process of investing in stocks. The effect of financial sophistication on stock
market participation is confirmed in a large number of alternative specifications and remains
unaffected when we employ a variety of robustness checks (see Van Rooij, Alessie and
Lusardi, 2007).
The fact that financial knowledge boosts stock ownership provides an opportunity to
exploit the risk premium on equity investments and might contribute to the positive effect of
financial literacy on net worth. This is true regardless of the fact that some households may in
fact be better off by not investing in the stock market due to excessive trading or a bad timing
of transactions as the evidence in the finance literature shows that the vast majority of
households investing in the stock market follow very passive investment strategies (see e.g.
Ameriks and Zeldes, 2004).
6.2 Financial sophistication and retirement planning
A second potentially important channel for wealth accumulation is that financial
sophistication is related to planning behavior. As an example, the model by Reis (2006)
distinguishes inattentive consumers who do not plan and do not accumulate wealth from those
who do plan and thereby accumulate savings. Empirical evidence supports the assertion that
planning affects wealth accumulation (Ameriks, Caplin and Leahy, 2003; Lusardi and
114
Mitchell, 2007a). Planning is an inherently complex task requiring advanced cognitive skills
and financial understanding. One needs to collect and process information from different
sources on current and future income and expenditures and calculate savings needs based
upon alternative scenarios. Thus, it is obvious that the effect of financial literacy on total net
worth might be related to planning capabilities.11 Indeed, Lusardi and Mitchell (2007b) report
convincing evidence of financial sophistication fostering thinking about retirement. In another
study, Lusardi and Mitchell (2008) document a positive relation between simple measures of
financial knowledge and more specific measures of retirement planning related to the
calculation of savings needs. In the following, we take these two approaches a step forward
by relating the more concrete definition of retirement planning to well-developed measures of
financial sophistication.
Our survey module contains a series of questions on retirement planning developed by
Lusardi and Mitchell (2006) and inserted in the 2004 wave of HRS. The precise wording of
the questions and variation therein depending on marital status and employment status are
reported in appendix B. The first question relates to the very first step in setting up a
retirement plan: ‘Have you ever tried to figure out how much your household would need to
save for retirement?’. Out of 1508 respondents 564 answered affirmatively and are labeled as
‘simple’ planners. The proportion of simple planners is comparable to the one found for US
households in HRS 2004, although the latter figure is based on a sample of older households.
Those respondents who answered ‘yes’ were given the next follow-up question: ‘Have you
developed a plan for retirement saving? The majority seems to have developed some sort of a
retirement savings plan as 161 plus 299 respondents answered ‘yes’ or ‘more or less’,
respectively. Out of this group of ‘serious’ planners, the large majority claims to have been
successful in the sense that 169 plus 250 respond ‘always’ or ‘mostly’ to the third question
‘How often have you been able to stick to this plan’. The proportion of simple, serious and
successful planners is roughly comparable to, albeit somewhat higher than, the findings for
US households in HRS 2004, although the latter is based on a sample of elderly households
(Lusardi and Mitchell, 2006). The weighted percentage of simple, serious and successful
planners in our sample equals 34.6, 27.6, and 25.1 respectively. 11 Even if people outsource much of the work to financial planners, they have to come up with a lot of information some of which is complex to retrieve and communicate (e.g. subjective information on their preferences and the uncertainty around the main scenario they foresee). At the same time, they have to be financially smart enough to understand the implications of their advice to judge whether these plans indeed fit their needs. Interestingly, a multivariate regression analysis reveals that financial sophistication does not exert an independent effect on the probability of consulting a financial intermediary. Illiterate households do however rely significantly more often on the advice of friends and acquaintances when making important financial decisions (results are available upon request).
115
Table 6. Retirement planning across demographics Weighted household percentages __________________________________________________________________ Percentage of planners ______________________________ Education Simple Serious Successful N ________________________ ________ ________ ________ ____ Primary 20.6 16.9 15.9 67 Preparatory intermediate voc. 37.3 27.6 25.1 345 Intermediate vocational 33.0 26.2 22.7 295 Secondary pre-university 33.1 26.6 23.1 207 Higher vocational 35.5 30.8 29.1 397 University 39.8 29.9 28.9 197 Pearson chi2(5) 9.50 3.37 4.75 p-value 0.09 0.64 0.45 __________________________________________________________________ Age Simple Serious Successful N ________________________ ________ ________ ________ ____ 21-30 years 24.8 18.5 14.9 179 31-40 years 30.0 23.0 21.8 306 41-50 years 34.6 27.1 24.8 333 51-60 years 45.4 36.7 34.0 311 61-70 years 34.8 28.4 25.3 217 71 years and older 34.4 28.9 27.0 162 Pearson chi2(5) 23.4 19.7 19.8 p-value 0.00 0.00 0.00 __________________________________________________________________ Gender Simple Serious Successful N ________________________ ________ ________ ________ ____ Female 32.6 26.5 24.4 674 Male 36.6 28.4 25.7 834 Pearson chi2(1) 0.42 0.03 0.02 p-value 0.52 0.86 0.88 __________________________________________________________________ Marital status Simple Serious Successful N ________________________ ________ ________ ________ ____ Single/divorced/widow 0.323 0.237 0.213 Married/living together 0.364 0.304 0.279 476 1032 Pearson chi2(1) 1.59 3.35 4.04 p-value 0.21 0.07 0.04 __________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
116
Descriptive statistics on retirement planning and demographics are reported in Tables
6 and 7. As expected, there is a strong correlation with age. The closer people get to
retirement the more likely they are to start considering their retirement needs. No differences
in planning activities between men and women come forward, while couples are more likely
to be successful in executing their plans. While there is not much evidence that planning is
related to education or basic literacy, there is a strong correlation with advanced financial
literacy. The proportion of planners in the most literate group is almost double the number for
households with the lowest level of financial understanding.
The relation between financial sophistication and simple retirement planning is
confirmed in a multivariate regression analysis including the same explanatory variables as
before (Table 8). We report OLS and IV regressions, as we are cautious of possible
simultaneity bias because one could become more financially educated in the process of
calculating savings needs, and developing and executing a retirement plan. The IV-
coefficients however suggest that the downward bias in the OLS coefficient due to imprecise
measurement of financial sophistication is more important than the effect of planning on
financial sophistication. A one standard deviation increase in financial sophistication
increases the probability to plan for retirement with more than 20 percentage points. Another
interesting result is the role of confidence. Those people who are very confident in their
financial capabilities are more likely to start making calculations on how much they need to
save for retirement purposes. This suggests that worries about their own financial skills and
capacity to handle complex retirement savings decisions withhold people from attempting to
calculate retirement savings needs and setting up plans.
Critics might argue that, in particular in the Netherlands, it is not clear that
sophisticated persons decide to save more for retirement when they compare the expected
retirement income with their spending needs.12 Informed people could as well come to the
conclusion that they are currently holding more wealth than necessary and adjust their savings
downward, since the Dutch pension system is known to be relatively generous and the vast
majority of employees save via mandatory defined benefit retirement plans with compulsory
contributions (Van Rooij, Kool and Prast, 2007). Research into these issues however shows
that the replacement rates provided in Dutch mandatory pension system are in many cases
below the expectations from employees and insufficient to provide in the desired old age
standard of living (Van Duijn, Lindeboom, Lundborg and Mastrogiacomo, 2008; Binswanger
12 Also for the US the conclusion - drawn in many studies - that retirement savings are insufficient is not undisputed (Scholz, Seshadri, and Khitatrakun, 2006).
117
and Schunk, 2008). This suggests that also in the Dutch system doing retirement calculations,
and subsequently developing targets for spending and savings might help people to overcome
problems of self-control and to improve their wealth position.
Table 7. Retirement planning and literacy Weighted household percentages __________________________________________________________________ Percentage of planners ______________________________ Basic literacy Simple Serious Successful N ________________________ ________ ________ ________ ____ 1 (low) 31.9 23.8 21.7 217 2 33.7 27.9 22.9 284 3 31.4 26.4 24.0 350 4 (high) 38.1 29.5 28.2 657 Pearson chi2(3) 1.95 0.94 3.62 p-value 0.58 0.82 0.31 __________________________________________________________________ Advanced literacy Simple Serious Successful N ________________________ ________ ________ ________ ____ 1 (low) 24.5 19.9 18.6 330 2 31.8 22.9 20.9 354 3 38.2 31.7 28.3 371 4 (high) 44.1 35.5 32.5 453 Pearson chi2(3) 32.6 22.9 20.6 p-value 0.00 0.00 0.00 __________________________________________________________________ Self-assessed literacy Simple Serious Successful N ________________________ ________ ________ ________ ____ 1 (very low) 53.4 44.1 44.1 9 2 33.3 17.8 15.0 56 3 21.2 17.3 16.2 137 4 26.7 20.3 16.1 366 5 37.0 30.7 28.2 499 6 45.7 37.7 36.1 355 7 (very high) 51.4 42.7 41.5 45 Do not know 17.6 10.2 10.2 31 Refusal 27.2 13.9 13.9 10 Pearson chi2(8) 48.6 43.6 49.9 p-value 0.00 0.00 0.00 __________________________________________________________________ Note: Percentages may not sum up to 100 due to rounding.
118
Table 8. Retirement planning and financial sophistication _____________________________________________________________________________ OLS IV _______________ _______________ Advanced financial literacy index 0.072*** (4.13) 0.25*** (2.66) Basic financial literacy index 0.031* (1.79) -0.026 (0.71) Age dummy (30<age<=40) 0.026 (0.43) 0.056 (0.89) Age dummy (40<age<=50) 0.084 (1.39) 0.097 (1.62) Age dummy (50<age<=60) 0.18*** (2.99) 0.17*** (2.77) Age dummy (60<age<=70) 0.16** (2.16) 0.15** (2.04) Age dummy (age>70) 0.052 (0.62) 0.056 (0.69) Intermediate vocational education 0.0029 (0.06) -0.026 (0.49) Secondary pre-university education -0.0081 (0.15) -0.062 (1.02) Higher vocational education -0.033 (0.74) -0.080 (1.57) University education 0.073 (1.31) 0.0064 (0.10) Male -0.061* (1.79) -0.11** (2.55) Married -0.032 (0.87) -0.017 (0.44) Number of children 0.017 (0.92) 0.022 (1.20) Retired 0.034 (0.54) 0.020 (0.32) Self-employed 0.0090 (0.13) -0.000095 (0.00) Ln(household income) -0.13 (0.05) 0.28 (0.09) Ln2(household income) 0.029 (0.12) -0.012 (0.04) Ln3(household income) -0.0013 (0.16) 0.000004 (0.00) High confidence in financial skills 0.14*** (3.35) 0.13*** (2.98) Low confidence in financial skills -0.048 (1.30) -0.021 (0.51) Risk aversion dummy 2 (low) 0.0085 (0.13) -0.022 (0.32) Risk aversion dummy 3 0.023 (0.34) 0.0034 (0.05) Risk aversion dummy 4 0.017 (0.27) 0.028 (0.43) Risk aversion dummy 5 0.017 (0.24) -0.0078 (0.11) Risk aversion dummy 6 -0.052 (0.85) -0.025 (0.38) Risk aversion dummy 7 (very high) -0.010 (0.17) 0.031 (0.48) Smoking: now and then -0.046 (0.69) -0.034 (0.48) Smoking: daily (1-20 cigarettes) 0.0100 (0.20) 0.017 (0.33) Smoking: daily (> 20cigarettes) -0.096 (1.30) -0.10 (1.28) Drinking: daily (> 4 glasses) -0.024 (0.37) -0.030 (0.46) Constant 0.061 (0.01) -1.20 (0.11) Observations 1003 1003 R-squared 0.07 -0.01 p-value test age=0 0.01 0.06 p-value test education=0 0.38 0.32 p-value test income=0 0.46 0.78 p-value test confidence=0 0.00 0.00 p-value test risk aversion=0 0.84 0.93 p-value test smoking, drinking=0 0.68 0.71 p-value Hansen J test 0.25 F-statistic first stage regression 13.0 p-value exogeneity test 0.06 _____________________________________________________________________________ Note: Absolute value of robust t-statistics in parentheses; *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is a 0-1 dummy indicating whether respondents have tried to calculate savings needs for retirement. The most risk tolerant, none smoking and moderately drinking (4 alcoholic drinks or less a day) respondents are in the reference group. The advanced literacy index has been instrumented using dummy variables indicating how much the respondent’s education was devoted to economics. The reference group consists of those respondents whose education was devoted a lot to economics.
119
6.3 Cost of ignorance
A question of major relevance for public policy decisions is whether the impact of
financial sophistication on net wealth positions is not only statistically significant but also
quantitatively meaningful, in other words whether financial literacy really matters in
economic terms. From the household’s point of view it is important as well to know whether
it is worthwhile to invest time, effort and financial resources in building up a high level of
financial sophistication. The regression results that document a positive and statistically
significant effect of financial literacy on wealth accumulation provide also a basis for some
simple calculations on the difference in net worth associated with different levels of financial
sophistication.
Table 9 reports the difference in net worth for individuals with lower and higher levels
of financial sophistication based upon our estimate for the advanced literacy coefficient.13 The
table shows that a small increase in financial sophistication from just below the level of an
average consumer to somewhat above the average, i.e. from the 45th to the 55th percentile of
its empirical distribution, increases net worth in expected terms by €11500. This certainly
constitutes a non-negligible number as about 20 percent of the households in our sample hold
lower levels of total net worth. Wealth effects for larger improvements along the literacy
distribution are even more substantial: the net worth difference associated with the 75th
percentile of the financial literacy distribution up from the 25th percentile equals over €81000.
Comparison with the median net worth level of about €120000 and the mean household net
worth of less than two hundred thousand euro makes clear that this type of wealth differences
are associated with big jumps in the relative wealth position. The 95%-confidence interval
surrounding the last estimate ranges from €11500 to €150600 reflecting that the estimate for
the financial literacy coefficient is surrounded by substantial uncertainty. The net worth
difference associated with an increase from the bottom to the top tail of the empirical financial
literacy distribution is estimated at over €200000. Note that while these calculations provide
crude proxies to have an idea of the relevance of literacy in economic terms, they do not take
into account possible wealth effects of changes in risk attitude or other personal
characteristics associated with higher levels of financial literacy.
Summarizing, while recognizing that our calculations provide crude approximations, it
is clear that from a public policy point of view the wealth effects of financial sophistication
are likely to be substantial. Also for households it seems attractive in terms of wealth holdings
13 In the calculations we use the coefficient and confidence interval for the effect of financial sophistication on wealth from the preferred IV-specification among the regressions in Table 4B (see column 3).
120
to invest in financial education insofar as these efforts boost financial skills. For the ultimate
impact on personal welfare though it makes a difference whether higher wealth holdings
come from improved wealth management leading to the avoidance of financial mistakes and
to higher portfolio returns or alternatively are the result of households being in a better
position to control their expenses. The two channels that we have highlighted (stock market
participation and retirement planning) suggest that both mechanisms are at work here. That
said it is important to realize that any effect of financial education on household wealth is not
instantaneous but needs time to materialize.
Table 9. Net worth differences associated with different levels of financial sophistication _______________________________________________________________________________ Improvement within financial literacy distribution (percentiles)
Simulated net worth difference (thousands of euro)
____________________________ ______________________________________________ From To Expected 95%-confidence interval _____ _____ __________ ____________________ 45 55 11,5 (1,6 - 21,4) 40 60 24,1 (3,4 - 44,8) 25 75 81,1 (11,5-150,6) 10 90 181,6 (25,8-337,5) 5 95 220,9 (31,3-410,5) 1 99 251,1 (35,6-466,5) _______________________________________________________________________________ Note: the expected net worth difference and its 95%-interval are derived from the estimate and its 95%-confidence interval for the coefficient on advanced financial literacy in the IV specification from Table 4B, keeping the values of all other variables unchanged.
7. Concluding remarks
Household financial skills, their effect on economic decisions and the prevalence of
financial mistakes have become an important topic in policy debates. It obvious that the
management of a wealth portfolio nowadays requires more sophisticated knowledge and skills
than say two or three decades ago. Not only have households become more and more
responsible for their individual welfare, but at the same time the landscape of financial
markets and products has dramatically changed; changes that have been characterized by a
vast increase in complexity and possibilities. To the best of our knowledge, this is the first
study on the relation between financial sophistication and wealth accumulation. Using explicit
measures for the level of basic cognitive financial ability and more advanced measures of
financial sophistication, we have documented empirical evidence of an independent positive
effect from financial sophistication on wealth accumulation. This effect of financial
121
sophistication on accumulated savings is robust across different specifications and continues
to hold if we control for many other wealth determinants.
We have highlighted evidence of two important channels that are likely to contribute
to this relation which is the fact that financially literate persons are more likely to invest in
stocks and have a higher propensity to plan for retirement. We argue that this is the result of
financial sophistication lowering the costs of collecting and processing information, and
reducing planning costs. Thereby it facilitates the execution of financial decisions and brings
down economic and psychological thresholds for participating in the stock market or
calculating retirement savings needs and developing retirement plans subsequently. In
addition, we have illustrated that the economic effects of changes in financial sophistication
are likely to be substantial. Our estimates suggest that even small difference in financial
sophistication are likely to be responsible for substantial differences in wealth holdings, but
this figure easily extends to over €80000 for larger differences in financial sophistication
(comparing the expected net worth difference associated with the 75th and 25th percentile of
the empirical financial literacy distribution).
Our study is complementary to the studies by Bernheim, Garret and Maki (2001), and
Bernheim and Garrett (2003) who have shown that financial education in the US (either at
high school or via seminars at the work place) exert a positive impact on savings, but could
not identify whether this effect runs via its influence on tastes for saving, via the provision of
information and the supply of commitment devices, through a broad improvement in financial
literacy and reduction of financial mistakes or works mainly via peer effects. The latter might
be the case if at least some participants of financial education programs have increased their
financial sophistication and neighbors, relatives, colleagues or others benefit via word-of-
mouth information or community effects. Our work shows that financial sophistication does
directly boost wealth accumulation, but we cannot infer from this result that the effect of
financial education programs indeed runs through an increase in financial literacy.14 For this
we need to separate the impact of several financial educating programs on financial ability
and knowledge from other channels.
An alternative to financial education could be to consider and stimulate initiatives
aiming to simplify complex decisions or to increase the transparency of markets and products.
Ironically, firms have less of an incentive to come up with more transparent and simple
14 Interestingly, a further analysis shows that peer effects might indeed play an important role in financial behavior especially for those with less financial sophistication as they are more likely to consult friends and relatives as their most important source of information for advice on financial decisions.
122
products the larger the part of the population with low financial sophistication (Gabaix and
Laibson, 2006). The idea is that firms might employ strategies to profit from less
sophisticated individuals even if this means that part of these gains are used to subsidize
financial sophisticated individuals who make optimal use of selling strategies to attract less
sophisticated, more inattentive consumers.
From a policy perspective, the benefits of higher financial sophistication are clear. Our
results show that financial sophistication leads to higher net worth levels, boosts the
participation in the stock market and increases the propensity to plan for retirement. These
effects are very welcome as they all contribute to consumers being well equipped to take
individual responsibility for their financial well being over the life cycle. An important issue
that is beyond the scope of this paper but certainly warrants more study is how and to what
extent financial sophistication can be stimulated and enhanced effectively.
123
References Agarwal, S., J. Driscoll, X. Gabaix, and D. Laibson, 2007, The age of reason: Financial decisions over the lifecycle, NBER Working Paper, 13191. Agnew, J., L. Szykman, S. Utkus, and J. Young, 2007, Literacy, trust and 401(k) savings behavior, Center for Retirement Research Working Paper, 2007-10, Boston. Alessie, R., A. Lusardi, and A. Kapteyn, 1999, Saving after retirement: Evidence from three different surveys, Labour Economics, 6, 277-310. Ameriks, J., A. Caplin, and J. Leahy, 2003, Wealth accumulation and the propensity to plan, Quarterly Journal of Economics, 118, 1007-1047. Ameriks, J., A. Caplin, J. Leahy, and. T. Tyler, 2007, Measuring self-control problems, American Economic Review, 97, 966-972. Ameriks, J., and S. Zeldes, 2004, How do household portfolio shares vary with age?, mimeo, Columbia Graduate School of Business. Barber, B., and T. Odean, 2000, Trading is hazardous to your wealth: The common stock investment performance of individual investors, Journal of Finance, 55, 773-806. Barber, B., and T. Odean, 2001, Boys will be boys: Gender, overconfidence, and common stock investment, Quarterly Journal of Economics, 116, 261-292. Barsky, R., F. Juster, M. Kimball, and M. Shapiro, 1997, Preference parameters and behavioral heterogeneity: An experimental approach in the Health and Retirement Study, Quarterly Journal of Economics, 112, 537-579. Benartzi, S., and R. Thaler, 2004, Save more tomorrow. Using behavioral economics to increase employee saving, Journal of Political Economy, 112, 164-187. Benjamin D., S. Brown, and J. Shapiro, 2006, Who is ‘behavioral’?: Cognitive ability and anomalous preferences, mimeo, Harvard University. Bernheim, D., 1995, Do households appreciate their financial vulnerabilities? An analysis of actions, perceptions, and public policy. In: Tax Policy and Economic Growth, American Council for Capital Formation, Washington, DC, 1-30. Bernheim, D., 1998, Financial illiteracy, education and retirement saving. In: O. Mitchell and S. Schieber (eds.), Living with Defined Contribution Pensions, University of Pennsylvania Press, Philadelphia, 38-68. Bernheim, D., D. Garrett, and D. Maki, 2001, Education and saving: The long-term effects of high school financial curriculum mandates, Journal of Public Economics, 80, 435-565. Bernheim, D., and D. Garrett, 2003, The effects of financial education in the workplace: Evidence from a survey of households, Journal of Public Economics, 87, 1487-1519.
124
Binswanger, J., and D. Schunk, 2008, What is an adequate standard of living during retirement?, CentER Discussion Paper, 82. Browning, M., and A. Lusardi, 1996, Household saving: Micro theories and micro facts, Journal of Economic Literature, 34, 1797-1855. Calvet, L., J. Campbell, and P. Sodini, 2007, Down or out: Assessing the welfare costs of household investment mistakes, Journal of Political Economy, 115, 707-747. Cagetti, M., and M. De Nardi, Wealth inequality: Data and models, NBER Working Paper, 12550. Campbell, J., 2006, Household finance, Journal of Finance, 61, 1553-1604. Chan, S., and A. Stevens, 2008, What you don’t know cannot help you: Pension knowledge and retirement decision-making, Review of Economics and Statistics, 90, 253-266. Christelis, D., T. Jappelli, and M. Padula, 2007, Cognitive abilities and portfolio choice, mimeo, University of Salerno. Christiansen, C., J. Joensen, and J. Rangvid, 2008, Are economists more likely to hold Stocks?, Review of Finance, 12, 465-496. Clark, R., and M. D’Ambrosio, 2008, Adjusting retirement goals and saving behavior: The role of financial education. In: A. Lusardi (ed.), Overcoming the Saving Slump: How to Increase the Effectiveness of Financial Education and Saving Programs, University of Chicago Press, forthcoming. Cocco, J., F. Gomes, and P. Maenhout, 2005, Consumption and portfolio choice over the life-cycle, Review of Financial Studies, 18, 490-533. Duflo, E., and E. Saez, 2003, The role of information and social interactions in retirement plan decisions: Evidence from a randomized experiment, Quarterly Journal of Economics, 118, 815-842. Duflo, E., and E. Saez, 2004, Implications of pension plan features, information, and social interactions for retirement saving decisions. In: O. Mitchell and S. Utkus (eds.), Pension Design and Structure: New Lessons from Behavioral Finance, Oxford, Oxford University Press, 137-153. Dohmen, T., A. Falk, D. Huffman, and U. Sunde, 2007, Are risk aversion and impatience related to cognitive ability?, IZA Discussion Paper, 2735. Fuchs, V., 1980, Time preference and health: An explorative study, NBER Working Paper, 539. Gabaix, X., and D. Laibson, 2006, Shrouded attributes, consumer myopia, and information suppression in competitive markets, Quarterly Journal of Economics, 121, 505-540.
125
Guiso, L., and T. Jappelli, 2005, Awareness and stock market participation, Review of Finance, 9, 537-567. Guiso, L., M. Haliassos, and T. Jappelli, 2002, Household portfolios, MIT Press, Cambridge. Guiso, L., T. Jappelli, and D. Terlizzese, 1996, Income risk, borrowing constraints, and portfolio choice, American Economic Review, 86, 158-172. Guiso, L., P. Sapienza, and L. Zingales, 2008, Trusting the stock market, Journal of Finance, forthcoming. Gustman, A., and T. Steinmeier, 2004, What People Don’t Know about their Pensions and Social Security, in W. Gale, J. Shoven and M. Warshawsky (eds.), Private Pensions and Public Policies, Washington, D.C.: Brookings Institution, 57-125. Haliassos, M., and C. Bertaut, 1995, Why do so few hold stocks?, Economic Journal, 105, 1110-1129. Heaton, J., and D. Lucas, 2000, Portfolio choice and asset prices: The importance of entrepreneurial risk, Journal of Finance, 55, 1163-1198. Hilgert, M., J. Hogarth, and S. Beverly, 2003, Household financial management: The connection between knowledge and behavior, Federal Reserve Bulletin, 309-322. Hubbard, R., J. Skinner, and S. Zeldes, 1995, Precautionary savings and social insurance, Journal of Political Economy, 103, 360-399. Hurd, M., 1989, Mortality risk and bequests, Econometrica, 57, 779-813. Hurd, M., 1990, Research on the elderly: Economic status, retirement and consumption and saving, Journal of Economic Literature, 28, 565-637. Hurst, E., and A. Lusardi, 2004, Liquidity constraints, household wealth and entrepreneurship. Journal of Political Economy, 112, 319–347. Kapteyn, A., R. Alessie, and A. Lusardi, 2005, Explaining the wealth holdings of different cohorts: Productivity growth and social security, European Economic Review, 49, 1361-1391. Kapteyn, A., and F. Teppa, 2002, Subjective measures of risk aversion and portfolio choice, mimeo, RAND. Kimball, M., and T. Shumway, 2006, Investor sophistication, and the participation, home bias, diversification, and employer stock puzzles, mimeo, University of Michigan. Laibson, D., 1997, Golden eggs and hyperbolic discounting, Quarterly Journal of Economics, 112, 443-477. Lusardi, A., 1999, Information, expectations, and savings for retirement. In: H. Aaron (ed.), Behavioral Dimensions of Retirement Economics, Washington, D.C.: Brookings Institution Press and Russell Sage Foundation, 81-115
126
Lusardi, A., 2003, Planning and Saving for Retirement, Working Paper, Dartmouth College. Lusardi, A., 2004, Saving and the effectiveness of financial education. In: O. Mitchell and S. Utkus (eds.), Pension Design and Structure: New Lessons from Behavioral Finance, Oxford University Press, 157-184. Lusardi, A., and O. Mitchell, 2006, Financial literacy and planning: Implications for retirement wellbeing, Pension Research Council Working Paper 1, The Wharton School. Lusardi, A., and O. Mitchell, 2007a, Baby boomers retirement security: The role of planning, financial literacy and housing wealth, Journal of Monetary Economics, 54, 205-224. Lusardi, A., and O. Mitchell, 2007b, Financial literacy and retirement planning: New evidence from the American Life Panel?, Working Paper, Dartmouth College. Lusardi, A., and O. Mitchell, 2008, Planning and financial literacy: How do women fare? American Economic Review, 98, 413-417. Modigliani, F., and R. Brumberg, 1954, Utility analysis and the consumption function: An interpretation of cross-section data. In: K. Kurihara (ed.), Post-Keynesian Economics, New Brunswick, New Jersey, Rutgers University Press, 388-436. Mandell, L., 2008, Financial education in high School. In: A. Lusardi (ed.), Overcoming the Saving Slump: How to Increase the Effectiveness of Financial Education and Saving Programs, University of Chicago Press, forthcoming. Moore, D., 2003, Survey of financial literacy in Washington State: Knowledge, behavior, attitudes and experiences, Technical report 03-39, Social and Economic Sciences Research Center, Washington State University. Organization for Economics Co-Operation and Development, 2005, Improving financial literacy: Analysis of issues and policies, Paris, France. Reis, R., 2006, Inattentive Consumers, Journal of Monetary Economics, 53, 1761-1800. Rosen, H., and S. Wu, 2004, Portfolio choice and health status, Journal of Financial Economics, 72, 457-484. Scholz, J., A. Seshadri, and S. Khitatrakun, 2006, Are Americans saving “optimally” for retirement?, Journal of Political Economy, 114, 607-643. Smith, B., and F. Stewart, 2008, Learning from the experience of OECD countries: Lessons for policy, programs, and evaluations. In: A. Lusardi (ed.), Overcoming the Saving Slump: How to Increase the Effectiveness of Financial Education and Saving Programs, University of Chicago Press, forthcoming. Staiger, D., and J. Stock, 1997, Instrumental variables regression with weak instruments, Econometrica, 65, 557-586.
127
Thaler, R., 1994, Psychology and savings policies, American Economic Review, 84, 186-192. Van Duijn, M., M. Lindeboom, P. Lundborg, and M. Mastrogiacomo, 2008, Pension plans and pension outcomes, who gets it right? An investigation on the congruence of expected retirement replacement rates, mimeo, Netspar, Free University of Amsterdam. Van Els, P., W. van den End, and M. van Rooij, 2004, Pensions and public opinion: A survey among Dutch households, De Economist, 152, 101-116. Van Rooij, M., A. Lusardi, and R. Alessie, 2007, Financial literacy and stock market participation, NBER Working Paper, 13565. Van Rooij, M., C. Kool, and H. Prast, 2007, Risk-return preferences in the pension domain: Are people able to choose?, Journal of Public Economics, 91, 701-722. Venti, S. and D. Wise, 1998, The cause of wealth dispersion at retirement: Choice or chance?, American Economic Review, 88, 185-191. Venti, S. and D. Wise, 2000, Choice, chance, and wealth dispersion, NBER Working Paper, 7521. Vissing-Jorgensen, A., 2004, Perspectives on behavioral finance: Does “irrationality” disappear with wealth? Evidence from expectations and actions. In: M. Gertler and K. Rogoff (eds.), NBER Macroeconomics Annual 2003, Cambridge, MIT Press.
128
Appendix A. Measuring literacy and confidence
Basic and advanced financial literacy
The construction of the basic and advanced literacy index is explained in detail in a previous paper
(Van Rooij, Lusardi and Alessie, 2007). In short, the basic literacy index follows from a factor
analysis based on five simple questions. For each question, we create a binary dummy equal to unity if
the respondent provides the correct answer. The five questions measure numeracy and the
understanding of economic concepts (related to the working of inflation and interest rates) that are
more or less necessary in day-to-day transactions. The index of advanced literacy is based on eleven
questions about more advanced concept like the understanding of stocks and bonds, the relation
between risk and return and the benefits of diversification. To do justice to the important role of do-
not-know answers, we have created two binary dummies for each question, measuring whether the
question is answered correctly, and whether the respondent indicated that he did not know the answer,
respectively. A factor analysis on these 22 dummies clearly points to one factor that adequately
describes the variation in responses. The procedure employed takes into account the fact that we have
used minor variations in wording for three out of eleven questions to test the sensitivity of responses to
these variations.
Overconfidence and underconfidence
At the beginning of our survey, we ask respondents to assess their own literacy. Table A1 reports the
exact wording of the question and the distribution of the responses. We have grouped the bottom three
categories and the top two categories from the 7-points response scale to retrieve four categories with
about equal size. We also divide the basic literacy index based on five simple economic questions over
four different groups and thereby try to mimic the size of the groups of the self-reported literacy
groups. This provides us with a relative ranking of self-reported literacy and one for measured basic
literacy. Those respondents that rank themselves higher than the rank we obtain for the basic literacy
score are labeled overconfident and those who are modest about their financial skills compared to the
actual measure of basic literacy are labeled underconfident. Both variables are binary dummies taking
the value unity if the respondent is overconfident or underconfident, respectively, and zero otherwise.
This way, we end up with 404 overconfident respondents, 599 underconfident respondents, 464
respondents with an equal ranking for actual and self-reported literacy, and 41 respondents with
missing information because they did not answer the self-assessed literacy question. The fact that we
obtain more relatively underconfident than overconfident persons is related to the fact that we are not
able to match the group sizes exactly, since the top category for basic literacy is relatively large,
containing the 677 respondents (out of 1508) who answer all five questions correctly.
129
Table A1 Self-assessed literacy Number and percentage of respondents ____________________________________________________ How would you assess your understanding of economics (on a 7-points scale; 1 means very low and 7 means very high)? ____________________________________________________ N % _____________ _____________ 1 (very low) 9 0.60 2 56 3.71 3 137 9.08 4 366 24.27 5 499 33.09 6 355 23.54 7 (very high) 45 2.98 Do not know 31 2.06 Refusal 10 0.66 _____________ _____________ Total 1508 100.00 ____________________________________________________
130
Appendix B. Wording of questions and construction of variables used in empirical work This appendix provides information on important variables used in the regression analysis. The squared brackets in the retirement planning questions indicate the different wording used depending on the marital status of the respondent and depending on whether the respondent is retired or not. Risk aversion To what extent do you agree or disagree with the statement ‘Investing in stocks is something I don’t do, since it is too risky’ (on a scale from 1 to 7, where 1 means ‘completely disagree’ and 7 means ‘completely agree’)?
Completely disagree Completely agree [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7 This provides us with a measure of risk aversion. The reference group in the empirical work consists of those respondents who disagree completely (category 1). Economics education How much of your education was devoted to economics? [ ] A lot [ ] Some [ ] Little [ ] Hardly at all [ ] Do not know [ ] Refusal The instrument variable economics education in the past is used in the regression analysis by including four dummy variables for the response categories ‘some’, ‘little’, ‘hardly at all,’ and ‘do not know/refusal’ respectively. The reference group consists of those respondents whose education was devoted ‘a lot’ to economics. Taste for saving Some people spend all their income immediately. Others save some money in order to have something to fall back on. Please indicate what you do with money that is left over after having paid for food, rent, and other necessities (on a scale from 1 to 7, where 1 means ‘I like to spend all my money immediately’ and 7 means ‘I want to save as much as possible’)?
I like to spend all my money immediately I want to save as much as possible [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7 The measure of taste for saving used in the regression analysis is constructed by grouping together the two lowest categories (very few respondents have chosen the lowest level), recoding the remaining six levels of taste for saving from 1 (quite low) to 6 (very high). The reference group in the empirical work consists of those respondents who like to spend all their money immediately (category 1).
131
Time preference People use different time horizons when they decide about what part of the income to spend, and what part to save. Which of the time horizons mentioned below is in your household MOST important with regard to planning expenditures and savings? [ ] The next couple of months [ ] The next year [ ] The next couple of years [ ] The next 5 to 10 years [ ] More than 10 years from now The reference group in the empirical work consists of those respondents who state that the most important time horizon is shortest, i.e. the next couple of months (category 1). Self-control Do you find it difficult to control your expenditures? Please indicate how difficult you find this (on a scale from 1 to 7, where 1 means ‘very easy’ and 7 means ‘very difficult’)?
Very easy Very difficult [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7 The measure of self-control used in the regression analysis is constructed by grouping together the two highest categories (very few respondents have chosen the highest level), recoding the remaining six levels of self-control from 1 (very easy) to 6 (quite difficult). The reference group in the empirical work consists of those respondents who find it very easy to control their expenditures (category 1). Bequest motive Please indicate which of the following four statements about parents leaving a bequest to their children would be closest to your own opinion about this? [ ] If our children would take good care of us when we get old, we would like to leave them a considerable bequest [ ] We would like to leave our children a considerable bequest, irrespective of the way they will take care of us when we are old [ ] We have no preconceived plans about leaving a bequest to our children [ ] We don’t intend to leave a bequest to our children [ ] None of the above-mentioned statements
132
Carefulness To what extent do you agree or disagree with the statement ‘I would describe myself as a careful person’ (on a scale from 1 to 7, where 1 means ‘completely disagree’ and 7 means ‘completely agree’)?
Completely disagree Completely agree [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7
[ ] Do not know [ ] Refusal
The measure of carefulness used in the regression analysis is constructed by grouping together the two lowest categories (very few respondents have chosen the lowest category), recoding the remaining six levels of carefulness from 1 (quite low) to 6 (very high). The few respondents that have chosen ‘do not know’ are added to the last category. The reference group in the empirical work consists of those respondents who strongly disagree with the statement that they are careful person (category 1). Precaution To what extent do you agree or disagree with the statement ‘When there is possible danger, I take many precautions’ (on a scale from 1 to 7, where 1 means ‘completely disagree’ and 7 means ‘completely agree’)?
Completely disagree Completely agree [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7
[ ] Do not know [ ] Refusal
The measure of precaution used in the regression analysis is constructed by grouping together the two lowest categories (very few respondents have chosen the lowest category), recoding the remaining six levels of precaution from 1 (quite low) to 6 (very high). The few respondents that have chosen ‘do not know’ are added to the last category. The reference group in the empirical work consists of those respondents who strongly disagree with the statement that they take many precautions (category 1). Thinking about retirement How much have you thought about retirement? [ ] A lot [ ] Some [ ] Little [ ] Hardly at all [ ] Do not know [ ] Refusal In the regression analysis, we use a dummy which takes the value 1 if respondents have thought ‘a lot‘ or ‘some’ about retirement, and 0 otherwise.
133
Simple planning [Have you [or your husband/wife/partner] ever tried\Did you [or your husband/wife/partner] try] to figure out how much your household would need to save yourself for [retirement?/ before you retired?] [ ] Yes [ ] No [ ] Do not know [ ] Refusal In the regression analysis, we use a dummy which takes the value 1 if respondents answered affirmatively and 0 otherwise. Serious planning [Have you\Did you] [or your husband/wife/partner] develop(ed) a plan for retirement saving? [ ] Yes [ ] More or Less [ ] No [ ] Do not know [ ] Refusal In the regression analysis, we use a dummy which takes the value 1 if respondents answered affirmatively and 0 otherwise. Successful planning How often [have you [and your husband/wife/partner] been\were you [and your husband/wife/partner]] able to stick to this plan: would you say always, mostly, rarely, or never? [ ] Always [ ] Mostly [ ] Rarely [ ] Never [ ] Do not know [ ] Refusal In the regression analysis, we use a dummy which takes the value 1 if respondents answered affirmatively and 0 otherwise Self-assessed literacy How would you assess your understanding of economics (on a 7-points scale; 1 means very low and 7 means very high)?
Very low Very high [ ] 1 [ ] 2 [ ] 3 [ ] 4 [ ] 5 [ ] 6 [ ] 7
[ ] Do not know [ ] Refusal
The index of self-assessed literacy used in the regression analysis is constructed by grouping together the two lowest categories (very few respondents have chosen the lowest level), recoding the remaining six levels of self-assessed literacy from 1 to 6 and excluding ‘do not know’ answers and ‘refusals.’
135
PAPER FOUR
CHOICE OR NO CHOICE: WHAT EXPLAINS THE ATTRACTIVENESS OF DEFAULT OPTIONS?
(Co-author: Federica Teppa)
137
Choice or No Choice:
What Explains the Attractiveness of Default Options?
Maarten van Rooij (De Nederlandsche Bank and Netspar) Federica Teppa (De Nederlandsche Bank and Netspar)*
November 2008
Abstract The default option in individual decision making has proved to be a major attractor in a large number of situations. Yet, direct empirical evidence on the reasons for the importance of the default is still lacking. We have devised a new module for the Dutch DNB Household Survey and the US RAND American Life Panel to identify potential explanations for default choices and to provide empirical evidence on their relative importance for retirement savings, organ donation, voting, having a will, and no-consent decisions in marketing. The use of survey data allows us to study the behavior of the entire population and to control for a rich set of personal characteristics, as well as for labor market status, income, and wealth. Our findings confirm that the default option plays a pivotal role in individual decision making in the Netherlands as well as in the US. Moreover, choice behavior seems to be driven by different reasons across different situations in both countries, with a particularly strong role for procrastination and financial illiteracy. In addition, we find an important role for social norms and peer effects explaining the deviation from default options in the Dutch data. Key words: Default Choices, Individual Decision Making, Procrastination, Inertia, Financial Literacy, Endorsement, Trust, Conformity, Status Quo Bias JEL Classification: D12, D80, C90 ∗ Maarten C.J. van Rooij, Economics & Research Division, De Nederlandsche Bank, P.O. Box 98, 1000 AB, Amsterdam ([email protected]), and Federica Teppa, Economics & Research Division, De Nederlandsche Bank, P.O. Box 98, 1000 AB, Amsterdam ([email protected]). We are grateful to Rob Alessie, Marcello Bofondi, Lans Bovenberg, Stefan Hochguertel, Lex Hoogduin, Clemens Kool, Peter Kooreman, David Laibson, Theo Nijman, Peter Schotman, Paul Sengmüller, Robert Slonim, Peter van Els, Raymond van Es, Nathanaël Vellekoop, Peter Vlaar and participants in the Netspar Conference on the Micro-economics of Ageing (Utrecht, November 2006), the French Economic Association Conference on Behavioral Economics and Experiments (Lyon, May 2007), the CEPR/Netspar European Challenges Conference (Zurich, October 2007), the Netspar Pension Day (Tilburg, October 2007), the 10th Annual DNB Research Conference on Behavioral Economics: Challenges to Policy Makers and Financial Institutions (Amsterdam, November 2007) and the Netspar Pension Workshop ‘Aging Insured’ (Utrecht, January 2008) for helpful suggestions and comments. A special thank goes to Arie Kapteyn and Arthur van Soest, for their comments and suggestions to improve upon the questionnaire and their help in using the RAND American Life Panel to collect US data. We thank Jeffrey Dominitz, Sandy Chien and Bas Weerman for their work in implementing the questionnaire in the US. We are also grateful to the staff of CentERdata, and in particular, Corrie Vis for their assistance in setting up the survey and the Dutch field work. The views expressed in this paper are those of the authors and do not necessarily reflect those of the institutions they belong to. Any remaining errors are our own responsibility.
138
1. Introduction
The role of default options in individual decision making is well documented in the
empirical and experimental literature. The polarization at the default is a persistent finding not
only in economics (e.g. pension savings, insurance), but also in other domains like organ
donation, phone marketing, and Internet privacy policies. These findings contradict the
predictions of neoclassical economics, as standard choice theory dictates that the framing of
choice problems, in particular the selection of one of the alternatives as the default, should be
irrelevant. As long as transaction or switching costs are small and preferences are well-
defined, the consumer will pick the option that maximizes her utility, irrespective of the
design of the choice problem.
The literature is less clear-cut on the reasons behind the attractiveness of default
options. Why is it that individuals are so much attracted to the default? Is there one obvious
reason or do different motivations play a role in different situations? Many potential
explanations have been suggested including inertia, procrastination, the interpretation of
defaults as endorsements, as well as choice overload and the complexity of choice problems.
Despite the great deal of attention devoted to the topic, a comparative study on the role of the
default in different settings seems to be missing. Nevertheless, these are important questions.
As an example, from a public policy perspective, it is relevant to know whether
nonparticipation in retirement plans or donor registration is a deliberate choice, and - if not -
whether it is the consequence of a lack of knowledge rather than of procrastination.
To answer these questions, we have designed a specific module for both the DNB
Household Survey in the Netherlands and the RAND American Life Panel in the US to elicit
information on personal traits and choices made in several situations with a default option.
The default is defined as the situation that occurs if an individual does not take any action. In
our empirical analysis, we take into account that this situation can be either the outcome of an
active decision process or a deferral choice, i.e. the result of not choosing or not taking any
action. We consider several choice domains including retirement savings, organ donation,
having a will, voting participation and no-consent decisions about phone and leaflet
commercial marketing. We provide empirical evidence on the relation between individual
choices and personal traits and background characteristics. To what extent respondents are
exposed to behavioral attitudes is identified by factors extracted from a principal component
analysis on a set of statements about individual behavior.
This paper contributes to the literature in a number of ways. First, it compares the role
of the default option across several domains. Second, it provides empirical evidence on the
139
relative importance of potential explanations for default choices. Third, the analysis is based
on survey data. Existing studies use either administrative data or field and laboratory
experiments with limited background information on the participants (see e.g. Madrian and
Shea, 2001, or Dhar, 1997). The intrinsic nature of such data sources prevents from
quantitatively testing the relevance of any of the potential reasons listed above. Moreover, the
use of survey data delivers a rather complete picture of default behavior, as the interviewed
people belong to the entire population distribution rather than to a particular sub-sample
(typically students or employees at selected firms). Fourth, the use of comparable Dutch and
US data enables a cross-country comparison.
Our main conclusions are the following. The default option plays an important in role
in many situations. Individual choices are driven by different determinants across domains,
but overall procrastination and financial illiteracy appear to be the most prevailing
explanations for default choices. This is true in the Netherlands as well as in the US, despite
differences in tradition, culture and institutions. In addition, we find that endorsement or
community effects play an important role in the Dutch data; individuals giving above average
weight to the opinion of others more often take decisions that are commonly viewed as being
part of good citizenship, as in the case of voting or the registration for organ donation.
These findings have important policy implications. Despite the standard theoretical
predictions, the default option turns out to be relevant for individual decision making. Thus,
policy makers need to be careful in framing choice situations as their design is not neutral.
Moreover, since the role of default is not driven by a unique determinant, the optimal design
of defaults should take this into account. On one hand, when agents stick to the default
because of procrastination, they may be better off with a design defaulting them into the
option that is the most appropriate for them. On the other hand, when agents do not choose
because of the complexity of the choice problem, education and information might be more
welfare improving, especially when individual preferences are heterogeneous (e.g. in the case
of pension savings). At the same time, increasing the simplicity of choice situations facilitates
active decision making.
The paper is organized as follows: In Section 2, we provide a review of the literature
on the role of default options in individual decision-making and potential explanations for
their attractiveness. In Section 3, we describe the data used in the empirical analysis for the
Netherlands, including the identification and validation of personal traits. In Section 4, we
report the descriptive statistics for choice behavior in situations with a no-action default. In
Section 5, we analyze the default choices and relate them to individual traits. In Section 6, we
140
replicate the empirical analysis with US data from the RAND American Life Panel. In
Section 7, we discuss the empirical results. In Section 8, we conclude with some final
remarks.
2. Literature
A rapidly growing literature (largely focused on the US) points out that, when taking
decisions, individuals rely on default options heavily. There are many event studies in
marketing research documenting the impact of framing on consumer decisions. Johnson,
Bellman and Lohse (2002) for example show that the default is relevant for Internet privacy
policies. Examining online permission for the addition to e-mail distribution lists for future
contacts, they find significant differences between opt-in and opt-out frames. Johnson,
Hershey, Meszaros and Kunreuther (1993) document the consequences of alternative choice
designs for car insurance by exploiting the variation in US state legislation. In Pennsylvania
by default insurance plans include the full right to sue for any auto-related injury with the
option to forego the full right in exchange for lower insurance premiums. In New Jersey car
drivers acquire a restricted right to sue unless they actively choose otherwise. Differences in
participation rates between the opt-out and the opt-in states were huge: 75 versus 20 percent,
even though the full right insurance option is not costless at all.
The design of organ donor registration might literally bring about differences between
life and death for those waiting for an organ donor transplant. Countries where everyone is
defaulted into organ donation unless he registers his unwillingness to be one have much more
potential organ donors than countries where nobody is a donor unless he explicitly signs a
consent statement. The effective consent rates range from well below half of the population in
the explicit consent countries to over 80 percent (and often close to 100 percent) in the
presumed consent countries (Johnson and Goldstein, 2003). Moreover, after controlling for
other determinants, actual donation rates as well appear to be considerably higher in countries
where citizens are defaulted into organ donation (Abadie and Gay, 2004).
Another influential area of research on the effect of default choices focuses on life
cycle savings behavior, especially in retirement plans. Madrian and Shea (2001) have
documented convincing evidence of a strong influence of plan design on savings choices.
They evaluate the consequences of default in a large US company in the health sector that
changed their opt-in 401(k) plan into an opt-out design. This change provides a sort of natural
experiment with new employees being enrolled automatically in a retirement savings plan
with a fixed contribution rate invested in a default money market fund, unless they explicitly
141
state other preferences for participation, contribution or portfolio investment. Prior to the
change, new employees were free to decide upon all these features of the retirement plan, but
only after an active decision to join. Instant participation rates rose significantly from 37 to 86
percent, with the vast majority contributing the default premium rate and investing all money
in the standard fund. Choi, Laibson, Madrian and Metrick (2004) extend the analysis to a
longer time horizon and find that the enrolment gap is still substantial after four years. Both
studies show that automatic enrolment is particularly successful raising the participation rate
of lower-pay employees. This suggests that default behavior and the sensitivity to framing in
choice situations might be related to personal characteristics.
The empirical evidence on the role of default in pension choices extends to other
counties than the US. Cronqvist and Thaler (2004) document investment behavior in Sweden,
where after a pension reform in 1999 employees had to decide how to invest part of their
pension premiums in private social security accounts (‘Premium Pension Funds’). One third
of the participants chose the default allocation despite the government urging them not to do
so. The proportion of default choices rose to 93 percent three years later, after the government
stopped its campaign. This illustrates that there might be an important role for information,
advertising and publicity campaigns surrounding choice occasions.
Bütler and Teppa (2007) show that default choices are relevant not only for pension
wealth accumulation but also for the decumulation phase. Some company pension funds in
Switzerland pay out the accrued employer pension savings as a lump sum upon retirement;
other funds transfer the total capital into a lifetime annuity. While both types of companies
offer the possibility to opt out the standard situation, Swiss pension fund participants
massively take the default option of their pension fund for granted.
The evidence on the reasons behind the attractiveness of default options is less
univocal. Samuelson and Zeckhauser (1988) introduce the term ‘status quo bias’ to describe
the tendency for individuals to stick either to their original choices or to the current situation.
The basic idea is that this preference is driven by loss aversion, i.e. individuals weigh losses
more strongly than profits (Kahneman and Tversky, 1979), and the fact that the status quo
serves as a reference point for their loss evaluation. Ritov and Baron (1992) claim that
individuals prefer inaction above action, regardless whether the status-quo is maintained or
not. This type of inertia is supported by Kahneman and Tversky (1982) and Landman (1987)
who document evidence of individuals regretting an unfortunate situation more if it is the
result of an active decision than if it happens because the person did not make an active
choice.
142
More recently, a number of studies have highlighted the relevance of procrastination
due to a lack of self-control, i.e. the tendency of people to postpone unpleasant tasks because
of a bias for immediate gratification. The underlying concept is that of individuals discounting
time inconsistently: their short term discount rate is smaller than the discount rate used for
decisions in the far future. One example of time inconsistent discounting is hyperbolic
discounting (Laibson, 1997), but it extends to broader classes of time preferences
(O’Donoghue and Rabin, 1999a, 2001).1
O’Donoghue and Rabin (2001) show that not only naïve but also more sophisticated
individuals might suffer from procrastination. Their model illustrates that providing non-
procrastinators with additional choices might induce procrastination in important tasks and
that the welfare costs of such behavior might be huge as in the case of insufficient pension
savings.2 The intuition is that individuals want to put effort in collecting information and
thinking about choices with major implications. Important choices as those related to
participating in a retirement savings plan thus require substantial short-term effort and costs
which might invoke procrastination in the optimistic view that this decision will be tackled in
the near future. This suggests that there is an additional role for complexity in relation to
cognitive ability in determining the attractiveness of default options.3 A high level of financial
sophistication for example reduces the costs of important financial choices and illiterate
individuals might show a higher aversion to taking these decisions. Indeed, Agnew and
Szykman (2005) provide experimental evidence of financially illiterate participants being
more likely to choose the default in complicated exercises.
The importance of financial literacy and advice-seeking also suggests that especially
individuals who are careful or take many precautions face high costs in making important
decisions as they are inclined to search for many sources of information and advice and think
at least twice before entering a new situation. The study by Kapteyn and Teppa (2002)
provides empirical evidence of the relevance of these attitudes for portfolio choices.
1 The idea of time-inconsistent discounting is not new however and goes back to the work of Strotz (1956), Phelps and Pollak (1968), Pollak (1968), and Akerlof (1991). 2 See O’Donoghue and Rabin (1999b) for an extensive discussion on how procrastination might have huge economic costs in terms of retirement savings. 3 Tversky and Shafir (1992) also show that when choices are difficult, i.e. when there is not one dominating alternative, it might be optimal to postpone the decision or alternatively go along with the default option to gather more information or search for alternatives. There is also experimental evidence that while basic choice theory suggests that increasing the number of choice options is always goods since the additional options may contain better alternatives, it in fact may prove to be demotivating and create dissatisfaction (Iyengar and Lepper, 2000). Indeed, empirical studies on asset allocation decisions provide examples of choice overload and participants looking for simplicity (see e.g. Iyengar and Kamenica, 2008, or Huberman and Jiang, 2006).
143
Another important motivation for the importance of default options is that the default
is seen as an advice or endorsement (Madrian and Shea, 2001; Beshears, Choi, Laibson and
Madrian, 2007). Individuals who are more sensitive to advice or in general relying more on
advice might also be inclined to go along with the default option. In case of 401(k) savings
plans the employer might be convinced that his employer wants the best for him, while in fact
also other arguments could play a role (e.g. pension costs or liability issues). Similarly,
experimental evidence confirms that people might perceive the way organ donation is
organized as a reflection of the policymakers’ preferences and the urge to participate
(McKenzie, Liersch and Finkelstein, 2006).
All these potential explanations for the attractiveness of default options are not
mutually exclusive but emphasize several aspects of choice behavior from a different
perspective (Beshears, Choi, Laibson and Madrian, 2007). In the next section, we identify to
what extent individuals are exposed to these types of behavior, i.e. we measure personal traits
which - compared to for example age, gender, and education - are less easily observed.
3. Data
We have collected information on individual choices in several situations with a
default option from the households participating in the DNB Household Survey (DHS). The
DHS, formerly known as the CentER Savings Survey, is an annual survey of about 2000
households in the Netherlands that started in 1993. In principle all household members aged
16 years and older are allowed to participate. The panel is run at Tilburg University by
CentERdata.4 In case of attrition, CentERdata recruits new participants to maintain the panel
size and to keep the panel representative on a number of relevant background characteristics
such as age, gender, income, education, and region of residence. The DHS dataset further
contains detailed information on employment status, pension arrangements, accommodation,
wealth, as well as health status and psychological concepts. The dataset thus provides the
opportunity to combine both economic and psychological aspects of financial behavior.
The module we have devised on default behavior was fielded in the weekend of June
2-6, 2006. Out of the 2467 panel members contacted, 1648 completed the questionnaire,
corresponding to a response rate of 66.8 percent. By merging our data with the annual DHS
survey, we are able to exploit the rich aforementioned information set. The age of the
respondents in our sample ranges from 16 to 91 years (mean age is 48.5); men and women are
4 More information on CentERdata, the CentERpanel and the DHS is available at their website (http://www.uvt.nl/centerdata/dhs).
144
equally represented (men account for 52.6 percent). As for household composition, 71.1
percent of the respondents are married or living with a partner, the others are single heads of
the household (22.9 percent) or children living with (one of) their parents (6.0 percent). Two
out of three respondents have children themselves. About one third of the respondents have a
college education (which includes vocational training in addition to university degrees), about
one third have an intermediate education level (secondary pre-university and intermediate
vocational), and about a third have a lower education level (primary and preparatory
intermediate vocational training). Overall, 19.6 percent of respondents are retired (including
early retirees), 49.5 percent are employees, and 3.7 percent are self-employed. The remainder
of the sample consists of individuals who are not retired and not working, including those
who are disabled or unemployed and those who follow an education program or take care of
the housekeeping.
3.1 Elicitation of individual traits
A novelty of this paper is that we link default choices to personal traits which are not
directly observable. We present the interviewees 17 statements on personal attitudes and
choices in real life situations that reveal information on individual traits that are expected to
be relevant for default behavior. The respondents are asked to indicate to what extent they
agree with each of the statements on a scale from 1 (‘totally disagree’) to 7 (‘totally agree’),
and they have the possibility to indicate that they ‘do not know’ or ‘refuse to answer’. The
statements have been presented in a random order to prevent any ordering effects in response
patterns.
Table A1 in appendix A reports the wording of these questions and the responses.
There are questions on whether and how people collect advice (Q1-Q3), the importance of
advice for their decisions (Q4-Q5), the role of the opinion of other people (Q6-Q9), whether
the interviewees tend to postpone tasks or decisions (Q10-Q12), whether they have a
preference for the status quo or no-change situation (Q13-Q14), as well as on carefulness and
precaution (Q15-Q16). In addition, we included a statement on financial literacy (Q17).5
5 Instead of inserting many different questions to measure financial knowledge and ability, we have included one question on self-assessed literacy that has proved to be a good proxy for more advanced measures of financial sophistication (Van Rooij, Lusardi and Alessie, 2007). In addition, the self-assessment of financial literacy might be more relevant for the respondents’ inclination to deviate from a default.
145
Response patterns reveal a high degree of heterogeneity among respondents. The number of
refusals and do not knows is limited.6
To summarize the information from the responses to the statements, we run a principal
component analysis for the 1509 respondents who provided an answer to the full set of
questions, i.e. we do not include respondents who filled in one or more ‘do not knows’ or
‘refusals’. It appears that the variation in the responses can be adequately captured by five
factors.7 Factor loadings measure to what extent each factor is correlated with the responses to
the original statements. For each statement we identify the factor with the highest correlation
(reported in Table A2 in appendix A). Reviewing these statements provides us with a
meaningful interpretation of the factors.8 The first factor is clearly related to the three
statements on procrastination. We label the second factor as trust as it measures to what
extent respondents gather and trust advice from family and friends.9 The third factor measures
inertia as it is related to the intensity in which people adhere to the status quo, possibly
because they want to carefully consider the alternatives before taking action. The fourth factor
measures to what extent people feel endorsed by others as it scores high on the statements
related to how important it is what other people say. The last factor measures self-assessed
financial literacy and the unwillingness to leave important decisions to somebody else.10
Based on the clustering in Table A2, we perform a principal component analysis on each
group of questions and extract principal component factors.11
3.2 Validation of individual traits
Table 1 reports the results of OLS regressions of the five factors on directly observable
information. Besides gender, age, education, and household composition, we include a
dummy for home ownership and quartile dummies for private household financial assets
6 We have experimented with additional statements, in particular on regret aversion. However, the number of ‘do not know’ and ‘refusal’ answers signaled that either the respondents did not have a strong opinion on these issues or that these questions were not fully clear to them. Therefore, we decided to exclude these additional statements from our analysis. 7 We retain factors with an eigenvalue that exceeds 1, i.e. those factors which explain a more than proportional part of the variation in responses. 8 The five factors are listed in order of relevance starting with the most important one, i.e. the one with the highest contribution to explaining the variance in response patterns of the original 17 statements. The cumulative proportion of variance that is explained by the five factors amounts to 53 percent. 9 The important role of trust or distrust in financial decision-making is highlighted by Guiso, Sapienza, and Zingales (2008), and Agnew, Szykman, Utkus and Young (2007). 10 This implies that the financial literacy factor extracted here does not have exactly the same meaning as in the other papers collected in this thesis. Nevertheless, we use this label to emphasize that it is the most important determinant of this factor. 11 One of the endorsement questions also loads on procrastination. We group it together with the endorsement questions as it appeals more strongly to this personal trait.
146
(bank and savings accounts and investments in stocks, bonds and mutual funds) and gross
personal income.12 This serves as a basic test for the validity of the identified personal traits
and provides additional information on whether they are not simply proxies for or closely
related to e.g. age or financial well-being.
Procrastination appears to be more prevalent among men and younger persons and,
surprisingly, the self-employed. Women, young respondents, those with children and couples
show more trust in the advice of other persons. Inertial behavior is more common among the
elderly and those with larger financial assets, i.e. we expect that these groups have a higher
likelihood of sticking to the current situation perhaps as a result of the wish to rethink changes
carefully before taking any action at all. Inertial behavior is less common for the highly
educated respondents and the self-employed. Apparently, for these groups any wish to
carefully consider alternatives is not a threshold for taking actions. Those with more
schooling might perceive lower costs in processing information necessary for comparing
alternatives and the self-employed are used to take many decisions. Endorsement is
negatively related to age; thus older cohorts seem less sensitive to the opinion of other people.
Financial literacy correlates strongly with gender, education, and financial well-being (income
and home ownership). Overall, the correlations found for the five personal traits we have
identified seem plausible and thereby underscore the interpretation given to the factors.
The ultimate validation of the individual traits lies in the association with factual
behavior. Panel members of the DNB Household Survey were given the chance to log in to
the survey website and fill in the questionnaire between Friday afternoon (5 pm) and Tuesday
midnight. CentERdata records the starting time and the duration of the interview. This
provides us with the opportunity to test whether actual response behavior is related to our
proxies for attitude.13 In particular, one could expect that those panel members who are more
inclined to procrastinate are also postponing the participation in the interview. We have
ranked the respondents according to their starting time and calculated the correlation with our
measure of procrastination. Indeed, our hypothesis is confirmed. The correlation is positive
(i.e. late participants in the survey are also those who procrastinate more) and strongly
significant.
12 For children living with their parents, we assume that their level of financial assets belongs to the bottom quartile. The regressions correlating the personal traits to background characteristics are based upon 1422 instead of 1509 observations because we lose some observations due to process of merging our survey with wealth and income information from other DHS modules. 13 We are indebted to Robert Slonim for this suggestion.
147
Table 1. Personal traits and individual background characteristics in the Netherlands _____________________________________________________________________________________ Procrastination Trust Inertia Endorsement Financial literacy ___________ _______ _______ __________ ______________ Male 0.33*** -0.36*** -0.083 -0.057 0.18*** (5.37) (5.92) (1.35) (0.92) (2.76) Age dummy (36-50) -0.18** -0.43*** 0.21*** -0.11 -0.025 (2.34) (6.34) (2.76) (1.42) (0.33) Age dummy (51-65) -0.28*** -0.56*** 0.33*** -0.31*** 0.079 (3.21) (7.24) (3.87) (3.58) (0.98) Age dummy (>65) -0.43*** -0.68*** 0.62*** -0.31** -0.066 (3.32) (5.29) (5.07) (2.40) (0.55) Intermediate education 0.016 0.024 -0.21*** -0.084 0.21*** (0.24) (0.38) (3.33) (1.18) (3.03) College education -0.070 0.053 -0.32*** -0.022 0.21*** (0.98) (0.77) (4.93) (0.30) (3.00) Employed 0.10 -0.092 -0.15* 0.043 -0.10 (1.16) (1.07) (1.82) (0.48) (1.20) Self-employed 0.32** 0.005 -0.30* -0.091 -0.21 (2.20) (0.04) (1.88) (0.67) (1.34) Retired 0.007 -0.036 -0.066 0.038 0.038 (0.07) (0.33) (0.65) (0.36) (0.34) Married -0.066 0.20*** -0.062 0.007 -0.036 (0.88) (2.75) (0.79) (0.09) (0.51) Has children -0.087 0.21*** 0.022 -0.043 -0.033 (1.17) (2.86) (0.31) (0.57) (0.47) Homeowner 0.041 -0.018 0.046 0.019 0.15** (0.60) (0.24) (0.65) (0.26) (2.03) Gross income quartile 2 -0.13 0.12 0.076 -0.046 0.28*** (1.50) (1.41) (0.88) (0.50) (3.04) Gross income quartile 3 -0.14 0.075 0.033 -0.067 0.36*** (1.39) (0.83) (0.33) (0.63) (3.36) Gross income quartile 4 -0.089 -0.005 -0.14 -0.080 0.59*** (0.76) (0.05) (1.35) (0.72) (5.21) Financial assets quartile 2 -0.089 0.17** 0.23*** 0.034 0.018 (1.11) (2.23) (2.90) (0.40) (0.23) Financial assets quartile 3 -0.096 0.11 0.40*** 0.12 -0.045 (1.18) (1.34) (4.98) (1.42) (0.56) Financial assets quartile 4 -0.16* 0.10 0.38*** 0.11 -0.027 (1.90) (1.19) (4.82) (1.25) (0.32) Constant 0.24** 0.20** -0.19* 0.21** -0.53*** (2.45) (2.30) (1.90) (2.15) (5.22) Observations 1422 1422 1422 1422 1422 R-squared 0.07 0.11 0.14 0.02 0.10 p-value test age = 0 0.00 0.00 0.00 0.00 0.26 p-value test education = 0 0.40 0.74 0.00 0.44 0.00 p-value test income = 0 0.38 0.28 0.025 0.91 0.00 p-value test finan. assets = 0 0.30 0.17 0.00 0.41 0.85 _____________________________________________________________________________________ Note: OLS estimation results; Standard errors are clustered at the household level: absolute value of robust t-statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%. The dependent variables are the output of the principal component analysis; The reference group for age contains respondents less than 36 years, for education respondents with low education and for income and wealth respondents in the bottom quartiles.
148
4. Decisions with a default alternative: empirical evidence
We explore actual choices in organ donation, voting participation, having a will,
consent to receive marketing, cancellation of subscriptions, retirement savings and early
retirement arrangements. While some of these domains are generally not thought of as
traditional default situations, each of them has a no-action alternative. Especially for voting
however there could also be another type of default, i.e. respondents might feel a moral
responsibility to vote. In addition, there can be peer pressure, i.e. the social norm in a
community might be that one ought to vote. Nevertheless, we can still learn about the
relevance of personal traits and attitudes for choice behavior. For example, the effect of
procrastination proves to be even more powerful if it induces people not to vote, despite
strong moral intentions and peer pressure. The same applies to organ donation.
4.1 Organ donation
Two systems of organ donation are used worldwide. In an opt-in system, individuals
are asked to register their willingness to become a donor. Countries that run the alternative
system assume that their citizens consent to organ donation unless they indicate otherwise and
explicitly opt out. In the Netherlands the former regime applies: people willing to donate their
organs have to record themselves in the donor register.
We have asked our respondents whether they think that in general people ought to be
prepared to be an organ donor, whether they are themselves willing to be an organ donor, and
whether they actually are organ donors, i.e. whether they are registered in the donor register
as being willing to act as organ donor. Table 2A reports the wording of the questions and the
responses. A large majority of the respondents (close to 70 percent) agrees that people ought
to be prepared to be an organ donor and that they are willing to be organ donors themselves
(differences in the responses to both questions are small). Yet, conditioning upon those who
are willing to act as organ donors, almost three out of ten persons are not registered. If the
government wants to improve upon this number (without changing the default of no-consent),
it is important to learn about the reasons why those 27 percent stick to the default.14
14 While 70 percent of our sample population has a positive attitude towards making their organs available, less than 50 percent (803 out of 1648) is registered as an organ donor. However, the official figure of organ donor registration is at most half of this percentage. This illustrates the fact that individuals participating in surveys are generally more socially involved (see also the discussion in Section 4.2).
149
Table 2A. Organ donation in the Netherlands __________________________________________________________________________________ Do you think in general
people ought to be prepared to be an organ donor?
Are you willing to be an organ donor?
Are you an organ donor, i.e. are you registered in the donor register as being willing to act as an organ donor?
_____________________ _____________________ ___________________________ Frequency Percentage Frequency Percentage Frequency Percentage ________ __________ ________ __________ _________ __________
Yes 1145 69.5 1121 68.0 803 71.6 No 321 19.5 320 19.4 299 26.7 Refusal 18 1.1 19 1.2 6 0.5 DK 164 10.0 188 11.4 13 1.2 ________ __________ ________ __________ _________ __________ Total 1648 100.0 1648 100.0 1121 100.0 __________________________________________________________________________________
Note: Default option in bold; DK = ‘I do not know’.
4.2 Voting participation
One of the distinguishing features of a democracy is universal voting: each individual
is entitled to participate in local or national elections, conditional on satisfying some legal
requirements, like age and nationality or residence. In most countries, including the
Netherlands, voting participation is a right and not a legal requirement. Thus, the default (no-
action) option in elections is not to vote. Respondents are asked to indicate whether they think
that in general everybody ought to vote, and whether they have voted themselves in the most
recent national (January 2003), European (June 2004) and local (March 2006) elections,
respectively. The reason to consider these three types of elections is that they could reveal
different information. Respondents might be more interested in national issues than in
discussions at the European level or they might feel more influential in local elections. In
addition, the amount and sort of publicity surrounding the different elections and the issues at
stake differ substantially.
Nine out of ten respondents indicate that in principle everybody should vote (Table
2B). A large number of them (ranging from 84 to 94 percent) did actually vote at the most
recent elections. The group of non-voters is relatively small, particularly in light of the fact
that the official statistics report substantially lower voting participation rates (80, 40 and 58
percent for the national, European and local elections, respectively).15 These findings
illustrate that individuals participating in (panel) surveys are generally more socially involved 15 Even if we use weights to correct for differences in the sample composition and population statistics regarding age, income, gender and education, this discrepancy does not vanish completely, at least not for the European and local elections. Weighted voting participation rates are 84 percent for the national elections, 74 percent for the European elections, and 77 percent for the local elections.
150
and more attached to the society which might lead to important differences in voting numbers
(Voogt and Saris, 2003). Note, however, that any self-selection along the dimension of social
involvement, while reducing the number of respondents that stick to the default, does not
affect the qualitative relations in our empirical analysis. If anything, as our sample is
intrinsically motivated to vote, explanations for sticking to the default instead of voting prove
to be really important.
In addition, we also cannot rule out the possibility that part of the discrepancy between
voting participation and official statistics relates to incorrect statements, i.e. non voting
respondents answering they did vote due to recall bias or social desirability bias. These
incorrect statements would show up in our empirical analysis as measurement error in the
dependent variable of voting participation regressions. However, insofar as the measurement
error is unrelated to the right hand side variables, the slope coefficients are still estimated
consistently. Moreover, the self-administered character of an internet panel not requiring
contact with an interviewer makes social desirable answers less likely than in personal
interviews (Chang and Krosnick, 2003).
Table 2B. Voting participation in the Netherlands __________________________________________________________________________________ Do you think in
general people ought to vote?
Did you vote last time for the national elections?
Did you vote last time for the European elections?
Did you vote last time for the local elections?
________________ ________________ ________________ ________________ Freq. Percent. Freq. Percent. Freq. Percent. Freq. Percent. ______ ______ ______ ______ ______ ______ ______ ______
Yes 1471 89.3 1378 93.7 1242 84.4 1279 87.0 No 114 6.9 87 5.9 214 14.6 191 13.0 Refusal 5 0.3 0 0.0 0 0.0 1 0.1 DK 58 3.5 6 0.4 15 1.0 0 0.0 ______ ______ ______ ______ ______ ______ ______ ______ Total 1648 100.0 1471 100.0 1471 100.0 1471 100.0 __________________________________________________________________________________ Note: Default option in bold; DK = ‘I do not know’.
4.3 Will, commercial leaflets, telemarketing and subscriptions
Table 2C reports choices related to having a will, the consent to receive marketing,
and the cancellation of subscriptions. A will or testament typically declares the destination of
a person’s belongings after her death or regulates the custody of children. A notary provides
advice, puts up the will and takes care of its execution. Having a will is neither a quick nor a
151
costless decision and not having a will is clearly the default. According to our data, some 60
percent of the respondents do not have a will.
In the Netherlands, as well as in a many other countries, a lot of companies and other
institutions massively send around unaddressed advertisements and publicity materials. What
is probably typical for the Netherlands is that people who are bothered by these mailings or
consider it is a waste of the environment may choose not to receive these commercial leaflets
by putting a ‘no/no’ (or a ‘no/yes’) sticker on their mailbox.16 The stickers are costless.
Sometimes they are distributed by local authorities, but they are always easy to order via
internet or by calling a special phone number. In addition, many companies and other
organizations hire call centers to approach potential customers by phone (often around dinner
time) to sell their products (telemarketing). As in many countries, it is possible in the
Netherlands to register yourself to get rid of these phone calls. Online registration is easy and
costless; one phone call or letter is sufficient to enter the register. Table 2C shows that even
though many households complain about the high number of superfluous commercial leaflets
they receive in their mailbox or annoying phone call during dinner time, only a small
proportion has undertaken any action to protect themselves from these marketing efforts.
Some 16 percent of the respondents have a sticker on their mailbox, and 12 percent has
registered to get rid of phone calls.
Table 2C. Will, commercial leaflets, telemarketing, subscriptions in the Netherlands __________________________________________________________________________________ Do you have a will? Do you have a
‘no/yes’ or a ‘no/no’ sticker on your mailbox?
Have you registered yourself in order not to receive telemarketing?
Are you thinking of canceling any subscriptions which are automatically continued?
________________ ________________ ________________ ________________ Freq. Percent. Freq. Percent. Freq. Percent. Freq. Percent. ______ ______ ______ ______ ______ ______ ______ ______ Yes 637 38.7 261 15.8 192 11.7 136 30.6 No 981 59.5 1349 81.9 1404 85.2 232 67.1 Refusal 9 0.6 6 0.4 6 0.4 0 0.0 DK 21 1.3 32 1.9 46 2.8 10 2.3 ______ ______ ______ ______ ______ ______ ______ ______ Total 1648 100.0 1648 100.0 1648 100.0 378 100.0 __________________________________________________________________________________ Note: Default option in bold; DK = ‘I do not know’. 16 The ‘no/yes’ sticker tells the mailman that the household living at this specific address does not want to receive commercial leaflets but does like to read the free local papers; the ‘no/no’ sticker stipulates that none of these are welcome.
152
Virtually all households have contracts or subscriptions for a fixed period of time,
often a year, which are automatically continued unless they are cancelled in time. Examples
include public transport cards, subscriptions to magazines, newspapers, and television-guides,
contracts for monthly participation in the lotto, membership of charity organizations, and
fitness clubs. Respondents are asked whether they intend to cancel any of them; and, if so,
why they have not cancelled these subscriptions yet. About a third of the respondents with
subscriptions actually intends to terminate one or more contracts but has not taken any action
yet (Table 2C).
4.4 Retirement savings
The pension system in the Netherlands consists of three pillars: the first pillar is a pay-
as-you-go state pension; the second layer consists of fully funded, privately provided pension
provisions; the third component is fully voluntary. Employees have hardly any discretion
about their first and second pillar arrangements that is if we disregard an indirect influence via
voting (potentially affecting the state pension) and via the negotiations of trade unions
(potentially affecting the company retirement plan). The state pension is a monthly benefit of
about €900 for single persons and the employer contributes part of the salary payments,
together with the company matching, to a pension fund that administers the company plan.
This way, over 90 percent of the Dutch employees saves compulsorily for their retirement
(Van Els, Van Rooij and Schuit, 2007).
The basic retirement choices that are available in the Dutch pension system are
whether to set apart additional savings via third pillar retirement savings products, or whether
to retire earlier than the regular retirement date. A third of the respondents has taken other
arrangements for their pension apart from the standard customary pension of the employer
(Table 2D). The others stick to the default, in this case meaning that they did not purchase
voluntary, often tax-deductible, pension products.
Until recently, early retirement schemes were often included in the - tax deductible -
compulsory second pillar savings scheme. There was a lot of heterogeneity among companies,
but within a company the freedom of choice was usually limited. Many companies run a
pension scheme with a retirement date before the age of 65 when Dutch citizens start
receiving the state pension. As of 2006, these early retirement schemes have become
unattractive due to a change in legislation. At the same time, the government introduced a
new savings vehicle for employees, the life cycle savings scheme (‘levensloopregeling’).
Employees are allowed to save up to 12 percent of their gross wage in a tax friendly way to
153
finance a future period of absence (e.g. early retirement or a sabbatical, parental, long care or
educational leave).
The publicity surrounding the introduction of the life cycle savings arrangement
emphasized the possible use for early retirement and trade unions and employers often
promoted it as the replacement of the existing early retirement schemes. Yet, the default is
non-participation. In fact, upon joining the life cycle savings arrangements employees have to
end their participation in another popular tax-favored savings plan for employees (the
‘spaarloonregeling’).17 Differences in tax facilities and many other characteristics complicate
the assessment of the relative attractiveness of both arrangements which also very much
depends on personal circumstances and preferences. However, if an employee is sure that he
wants to retire before the age of 65, it is certainly attractive to join the life cycle arrangement.
Yet, while the vast majority of employees plans to retire early, the participation rate in the life
cycle savings arrangement is limited to 8 percent (Table 2D). In the following, we will
basically interpret the life cycle savings account as an early retirement vehicle, since the
majority of participants intends to use this instrument for early retirement and to a lesser
extent for parental leave or a sabbatical period (Van Els, Van Rooij and Schuit, 2007).
Table 2D. Voluntary retirement and life cycle savings in the Netherlands __________________________________________________________________________________ Do you have other arrangements for your
pension apart from the standard customary pension you build up through your employer?
Do you participate in a life cycle savings arrangement (‘levensloopregeling’)?
_____________________________________ ______________________________ Frequency Percentage Frequency Percentage ________ ________ ________ ________ Yes 543 33.0 71 7.5 No 954 57.9 831 88.0 Refusal 18 1.1 6 0.6 Do not know 133 8.1 36 3.8 ________ ________ ________ ________ Total 1648 100.0 944 100.0 __________________________________________________________________________________ Note: Default option in bold. 4.5 The role of default options
The descriptive tables make clear that the default is a popular choice in many areas of
individual decision-making. The default option seems to attract the majority in domains
17 Currently, employees are allowed to save up to €613 out of their gross income per year without paying income or wealth taxes. After four years, the employee is free to spend these savings. Alessie, Hochguertel and Van Soest (2006) discuss this savings vehicle in more detail.
154
where the decision requires some additional financial skills (retirement savings) or in
situations where the marginal disutility associated to postponing the decision is lower (getting
rid of commercial leaflets, subscriptions or telemarketing). This is the case even in a non-
economic domain like organ donation where socially involved panel participants can be
expected to deviate more often from the default.
In interpreting these findings, one should take into account that the investigated
domains are heterogeneous. Organ donation, for example, is a reversible decision, potentially
driven by moral or religious convictions. Voting occurs at fixed dates and is an irreversible
but recurring action. Having a will is a reversible choice, but involving non negligible costs.
Getting rid of commercial leaflets or telemarketing is also a reversible decision, but by far
much less costly. Cancellations of periodical subscriptions are subject to deadlines. Finally,
voluntary (early) retirement saving is a continuous, dynamic choice, certainly requiring some
additional specific financial expertise. Obviously, these different properties might have
different implications for the decision making process. Below, we search for an explanation
for the variation in individual choices in these heterogeneous situations using the personal
traits identified in Section 3.
5. What explains the attractiveness of default options?
The purpose of this study is to investigate the explanation for default behavior, or
more precisely to identify the determinants that make a choice more likely only because it is
framed as the default option. Therefore, it is important to discriminate between two cases.
Sticking to the default option might be either the consequence of a lack of choice or the
outcome of an active decision process after careful consideration of the alternatives.
Respondents in the latter group would opt for the same option when the choice would be
framed differently. The former group is the one we are interested in. To make sure that the
regressions reported in this section concentrate on explaining default behavior, we therefore
exclude those observations where it is obvious that the default alternative coincides with the
preferred choice.
In the case of organ donation and voting we start conditioning our sample on those
respondents who are prepared to be an organ donor and those who agree that basically
everyone who is eligible ought to vote, respectively. Other selections are based upon the
responses to the question why the respondent did not deviate from the default option. In the
case of organ donation, this means that we also exclude those respondents who indicate that
they are not eligible, have instructed their family what to do, or have not registered because of
155
skepticism about the organ donation procedures. In studying the relation between individual
characteristics and not having a will, we exclude those respondents who have no children and
indicate that they do not have a will due to the absence of assets.18 Investigating the default of
not taking action to get rid of commercial leaflets or phone marketing, we disregard the group
who states that they find this kind of marketing useful; a group that is quite large in the case
of commercial leaflets. In the subscriptions domain, we limit our analysis to those people who
have subscriptions and indicate that they are thinking of canceling a subscription but did not
do so for other reasons than that they just made this decision and had to respect the terms of
cancellation. In analyzing what type of respondents do not have voluntary pension savings,
we exclude those who are retired or claim to have other assets making additional pension
savings unnecessary. In analyzing the participation in life cycle savings arrangements we
consider employees, excluding those who state that it is more attractive to save otherwise.
5.1 Personal attitudes and default choices
We have run probit regressions for each of the choice situations discussed in Section 4
based upon the respondents selected as explained above. The dependent variable takes the
value 1 if respondents choose the default option and the value 0 otherwise. Tables 3A and 3B
report the results in terms of marginal effects of a probit regression on the personal traits
identified in Section 4. A positive sign implies that the higher the degree of the corresponding
explanatory variable, the higher the probability of sticking to the default option. In most
regressions the personal traits clearly contribute to the explanation of default behavior.19
Procrastination and financial literacy seem to matter in many domains (mostly at the 1
percent significance level). The more individuals procrastinate, the higher the probability of
not being an organ donor, not voting, not having a will, not canceling subscriptions, and not
participating in the life cycle savings plan. The strongest effect is found in the subscription
domain: a one unit increase in procrastination20 reduces the probability of canceling
subscriptions by almost 10 percentage points. A similar increase in procrastination reduces
the probability of being an organ donor and having a will by approximately 4 percentage
points, and the probability of participating in the life cycle plan by 6 percentage points. A
18 Admittedly, the selection of respondents who consciously have decided not to have a will is m ore difficult than in other domains as a will can serve two main purposes regarding financial matters as well as taking care of the custody over young children. The results reported below do not change qualitatively however when the selection is based on financial assets only, irrespective of the presence of kids. 19 Only in the commercial leaflet regression the joint significance is far from the standard significance thresholds. 20 The five factors or personal traits are normalized, i.e. all have mean 0 and standard deviation 1.
156
smaller effect is found for voting behavior: the effect ranges from a 3 percentage points lower
probability of voting for the European elections to 1 percentage point for elections for the
national Parliament which by far attracts the most publicity and public interest.
Table 3A. Default choices and personal traits in the Netherlands __________________________________________________________________________________ Organ
donation Voting national
Voting European
Voting local
Will
__________ __________ __________ __________ __________ Procrastination 0.039*** 0.014*** 0.029*** 0.020** 0.042*** (3.11) (2.86) (2.85) (2.12) (2.81) Trust -0.014 0.014** -0.009 0.010 0.012 (1.02) (2.55) (0.94) (1.17) (0.76) Inertia 0.008 -0.007 -0.007 -0.016* -0.051*** (0.58) (1.37) (0.66) (1.73) (3.34) Endorsement -0.034** -0.018*** -0.012 -0.022** 0.018 (2.45) (2.90) (1.08) (2.28) (1.11)
-0.022 -0.020*** -0.012 -0.008 -0.054*** Financial literacy (1.63) (3.99) (1.34) (0.84) (3.56)
Observations 915 1369 1362 1375 1292 Pseudo R-squared 0.02 0.06 0.01 0.02 0.02 p-value test coeff. = 0 0.00 0.00 0.03 0.02 0.00 __________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, and 0 otherwise. Table 3B. Default choices and personal traits in the Netherlands __________________________________________________________________________________ Commercial
leaflets Phone
marketing Subscription Voluntary
pension savings Life cycle savings
_________ _________ __________ ______________ __________ Procrastination -0.003 -0.005 0.10*** -0.018 0.056** (0.16) (0.51) (3.67) (1.00) (2.51) Trust 0.009 0.005 -0.038 0.071*** 0.024 (0.47) (0.47) (1.50) (3.51) (0.95) Inertia 0.028 0.007 0.026 0.032 -0.007 (1.31) (0.72) (0.99) (1.61) (0.30) Endorsement -0.051** -0.006 0.042 -0.043** -0.041* (2.35) (0.70) (1.63) (2.20) (1.77)
-0.006 -0.033*** -0.007 -0.075*** 0.003 Financial literacy (0.29) (3.54) (0.27) (3.68) (0.12)
Observations 639 1463 415 808 302 Pseudo R-squared 0.01 0.01 0.05 0.03 0.03 p-value test coeff. = 0 0.22 0.01 0.00 0.00 0.09 __________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, and 0 otherwise.
157
The higher the degree of financial literacy, the higher the probability of voting for
national elections, having a will, getting rid of phone marketing, and having additional,
voluntary pension savings schemes. The marginal effects are largest for the supplementary
retirement savings (8 percentage points), and lowest for voting participation (2 percentage
points). Also in the domains where the effect of financial literacy is not significant, more
financially literate people tend to deviate from the default. The only exception is the life cycle
savings arrangement. 21There is also an important but unexpected role for the endorsement
factor. Our prior was that respondents that score high on this factor could interpret the default
option as an endorsement or implicit recommendation from the company on pension savings
(in line with the evidence on automatic enrollment plans in the US (Madrian and Shea, 2001))
or from the government on organ donation (McKenzie, Liersch, and Finkelstein, 2006). The
implication is that we would expect a positive association with the probability to choose the
default option in our regression. Instead, we find that these respondents more often deviate
from the default in the organ donation, pension and other domains.
To shed light on this relationship, we go back to the questions determining the
endorsement factor. These questions all relate to what other people say, suggesting that this
factor could as well measure a community effect, i.e. to what extent the respondent is
influenced by what friends, neighbors and colleagues think, say or do. Those who are more
sensitive to peer opinion are more likely to deviate from the default when the alternative is
commonly thought off as a good deed as is the case in e.g. organ donation and voting. This
interpretation is in line with the evidence on the importance of social interactions documented
in the literature for participation in retirement plans (Duflo and Saez, 2002, 2003) and the
stock market (Hong, Kubik and Stein, 2004; Brown, Ivkovic, Smith and Weisbenner, 2008).
It is also consistent with the implications of the theory of conformity by Bernheim (1994)
whose model shows how status and social interactions explain individuals behaving conform
perceived social norms. The interpretation of the endorsement factor as a measure of peer
pressure and conformity explains why some individuals are in a number of situations more
likely to deviate from the default option (e.g. organ donation, voting and retirement savings).
The role of trust and inertia seems to be less important for default decisions. Trusting
individuals are less likely to enter into additional retirement savings products; they are
21 We cannot draw strong conclusions from this as, besides being insignificant, the coefficient follows from a regression based on a relatively low number of observations. This said, it could indicate that literate employees do not assess the life cycle savings arrangement attractive enough e.g. because of the restrictions of these savings plans (tax advantages for example are foregone when the employee changes his mind and does not want to retire early or use the savings account for another period of absence).
158
apparently confident that the state pension plus the mandatory pension savings will make up
an adequate retirement benefit. The fact that inertia is positively related to having a will might
well be explained by the role of carefulness. Instead of postponing the decision to take a will
as a result of sustained deliberations, careful and precautious persons might be more
motivated to take care of survivors in the case of an unfortunate event.
5.2 Pooling personal attitudes and individual background characteristics
In the previous section, we have focused on the relation between personal attitudes and
default behavior. Now, we extend the set of control variables with personal background
characteristics (gender, age, level of education, job status, household composition, home
ownership, gross personal income, and household financial assets). Tables 4A and 4B report
the probit regression results. The personal background information contributes significantly to
the explanation of the observed choices. This is to be expected as benefits of deviating from
the default alternative are often related to background characteristics, e.g. having a will is
more likely for households with many real or financial assets.
Compared to the previous estimates, a striking difference is that procrastination is no
longer relevant for voting participation. This might be related to the age effect, as age appears
to be positively related to the likelihood of voting, whereas Table 1 shows that the elderly
procrastinate less. More interestingly, the other coefficients of the personal attitude variables
measuring procrastination, financial literacy and the effect of conformity remain by and large
unchanged. The level of significance is sometimes reduced, but the total number of
observations is also lower due to missing information for individual background variables
thereby decreasing the efficiency of the estimates. In addition, the weakening of financial
literacy might also be related to the fact that it served as a proxy for income and wealth in
previous regressions. Across the board, however, the estimation results confirm that
procrastination, financial illiteracy and conformity are important determinants of choice
behavior in the Netherlands not only in decisions of relatively minor relevance but also in
situations with a potentially huge impact on personal wellbeing (saving for (early) retirement)
or the wellbeing of others (having a will, organ donation).
Turning to the background characteristics that appear most relevant for individual
choice-making, gender significantly affects in particular organ donation, voting behavior (at
the national elections), and having a will. Compared with women, men have a 7 percentage
points lower probability to have filled in the organ donor registration form, a 3 percentage
points lower probability to vote, and a 10 percentage points lower probability to have a will.
159
Table 4A. Default choices and personal traits plus background characteristics in the Netherlands ____________________________________________________________________________________________________ Organ
donation Voting national Voting
European Voting local
Will
_____________ ___________ ___________ ___________ ___________ Male 0.073** 0.031*** 0.033 0.031 0.096** (2.43) (3.44) (1.48) (1.56) (2.44) Age dummy (36-50) 0.020 -0.017* -0.071*** -0.058*** -0.094* (0.50) (1.74) (2.91) (2.67) (1.70) Age dummy (51-65) 0.033 -0.028** -0.12*** -0.078*** -0.14** (0.76) (2.38) (4.22) (2.99) (2.42) Age dummy (>65) 0.082 -0.039*** -0.11*** -0.10*** -0.31*** (1.14) (2.85) (2.74) (3.31) (3.84) Intermediate education -0.048 -0.019** -0.039* -0.020 0.023 (1.48) (2.10) (1.74) (0.93) (0.54) College education -0.030 -0.026** -0.072*** -0.052** 0.017 (0.85) (2.19) (2.86) (2.14) (0.40) Employed -0.083* -0.008 -0.043 -0.014 0.042 (1.80) (0.65) (1.45) (0.52) (0.81) Self-employed -0.12** 0.019 -0.065 0.030 -0.11 (2.19) (0.69) (1.33) (0.58) (1.22) Retired -0.096** 0.008 -0.040 0.014 -0.039 (2.08) (0.46) (0.93) (0.40) (0.63) Married 0.005 -0.009 -0.012 -0.024 -0.066 (0.14) (0.82) (0.45) (0.97) (1.34) Has children -0.038 -0.023* -0.001 -0.014 0.031 (1.06) (1.85) (0.05) (0.56) (0.59) Homeowner -0.043 -0.007 -0.059** -0.018 -0.35*** (1.24) (0.78) (2.40) (0.80) (7.94) Gross income quartile 2 -0.016 -0.023** -0.008 -0.022 0.027 (0.38) (2.41) (0.25) (0.84) (0.51) Gross income quartile 3 -0.012 -0.020* -0.005 -0.013 -0.001 (0.24) (1.73) (0.14) (0.40) (0.02) Gross income quartile 4 -0.002 -0.029** 0.036 -0.038 -0.054 (0.04) (1.99) (0.88) (1.16) (0.80) Financial assets quart. 2 0.011 -0.009 -0.042* -0.028 -0.075 (0.28) (0.91) (1.65) (1.12) (1.39) Financial assets quart. 3 -0.062 -0.003 -0.035 0.003 -0.020 (1.56) (0.26) (1.37) (0.13) (0.36) Financial assets quart. 4 0.028 -0.018 -0.044 -0.019 -0.10* (0.68) (1.52) (1.57) (0.68) (1.84) Procrastination 0.044*** 0.004 0.009 0.004 0.030* (3.44) (0.88) (0.85) (0.44) (1.74) Trust -0.004 0.011** -0.018* 0.003 0.009 (0.29) (2.28) (1.71) (0.35) (0.50) Inertia 0.000 0.000 0.010 -0.007 -0.034* (0.02) (0.01) (0.94) (0.66) (1.87) Endorsement -0.037*** -0.016*** -0.010 -0.020** 0.017 (2.64) (3.36) (0.91) (2.20) (0.91) Financial literacy -0.018 -0.012*** -0.010 -0.004 -0.051*** (1.36) (2.95) (1.06) (0.37) (2.86) Observations 869 1285 1278 1291 1218 Pseudo R-squared 0.07 0.21 0.09 0.07 0.15 p-value test age = 0 0.73 0.02 0.00 0.00 0.00 p-value test education= 0 0.33 0.04 0.02 0.10 0.86 p-value test income = 0 0.97 0.06 0.51 0.61 0.43 p-value test fin. assets= 0 0.08 0.43 0.29 0.54 0.18 ____________________________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, and 0 otherwise; The reference group for age contains respondents less than 36 years, for education respondents with low education and for income and wealth respondents in the bottom quartiles.
160
Table 4B. Default choices and personal traits plus background characteristics in the Netherlands ____________________________________________________________________________________________________ Commercial
leaflets Phone
marketing Subscriptions Voluntary pension
savings Life cycle savings
________________________ __________ ________________ _________ Male 0.008 -0.020 0.12* -0.041 0.030 (0.17) (1.09) (1.87) (0.85) (0.66) Age dummy (36-50) -0.15** -0.074** 0.041 -0.041 0.031 (2.00) (2.50) (0.52) (0.73) (0.58) Age dummy (51-65) -0.13 -0.009 -0.059 -0.17*** -0.003 (1.52) (0.28) (0.65) (2.68) (0.05) Age dummy (>65) -0.32*** -0.012 -0.16 (3.01) (0.29) (1.08) Intermediate education -0.10* 0.015 -0.057 -0.074 0.044 (1.78) (0.70) (0.77) (1.45) (0.95) College education -0.25*** 0.040* -0.015 -0.049 0.13*** (4.22) (1.74) (0.21) (0.85) (2.66) Employed 0.077 -0.022 -0.22*** -0.11* 0.022 (1.09) (0.72) (2.60) (1.85) (0.30) Self-employed 0.20** -0.000 -0.085 -0.053 (2.20) (0.00) (0.61) (0.50) Retired 0.16** -0.026 -0.021 (2.06) (0.73) (0.18) Married 0.14** 0.026 -0.020 -0.079 0.042 (2.31) (1.02) (0.28) (1.41) (0.79) Has children 0.23*** 0.044* -0.079 0.011 0.048 (3.74) (1.78) (1.10) (0.21) (0.85) Homeowner -0.037 -0.040* -0.017 -0.11** -0.12*** (0.68) (1.71) (0.24) (2.31) (2.99) Gross income quartile 2 0.054 0.002 0.091 -0.072 -0.29** (0.73) (0.06) (1.05) (1.09) (2.48) Gross income quartile 3 0.089 0.022 0.070 -0.20*** -0.40*** (1.11) (0.64) (0.68) (2.76) (2.96) Gross income quartile 4 0.11 0.000 0.092 -0.29*** -0.48*** (1.23) (0.00) (0.86) (3.66) (3.15) Financial assets quart. 2 -0.013 -0.030 -0.099 -0.069 -0.034 (0.17) (1.07) (1.25) (1.21) (0.64) Financial assets quart. 3 0.003 -0.038 0.086 -0.088 -0.074 (0.04) (1.35) (1.11) (1.54) (1.25) Financial assets quart. 4 -0.064 0.022 0.005 -0.15** -0.16** (0.93) (0.73) (0.07) (2.46) (2.19) Procrastination -0.024 0.002 0.097*** 0.002 0.078*** (1.12) (0.26) (3.41) (0.07) (4.34) Trust -0.004 0.005 -0.022 0.050** 0.026 (0.18) (0.56) (0.75) (2.21) (1.19) Inertia 0.028 0.009 0.022 0.029 0.009 (1.19) (0.94) (0.71) (1.29) (0.42) Endorsement -0.055** -0.006 0.029 -0.033 -0.060*** (2.29) (0.65) (1.02) (1.56) (2.73) Financial literacy -0.014 -0.027*** -0.025 -0.047** 0.023 (0.56) (2.76) (0.87) (2.07) (1.13) Observations 599 1378 389 751 277 Pseudo R-squared 0.11 0.05 0.08 0.15 0.21 p-value test age = 0 0.01 0.03 0.43 0.01 0.68 p-value test education=0 0.00 0.21 0.71 0.35 0.02 p-value test income = 0 0.65 0.76 0.75 0.00 0.02 p-value test fin.assets=0 0.68 0.12 0.12 0.10 0.18 ____________________________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, and 0 otherwise; The reference group for age contains respondents less than 36 years, for education respondents with low education and for income and wealth respondents in the bottom quartiles.
161
Age is significant in seven out of ten cases. Older respondents are more likely to vote,
to have a will, and to have taken action to prevent them from receiving commercial leaflets.
Age is also related to voluntary pension savings. Older generations (not including those who
are already retired) have more often put additional money aside for their pension.
The level of education turns out to be jointly significant for voting, commercial
leaflets, and life cycle savings. The higher the education level, the higher the likelihood of
voting and getting rid of commercial leaflets and the lower the probability to join the life
cycle savings scheme.
Another control that contributes significantly to the explanation of choice behavior in
several domains is home ownership. Home ownership is among others relevant for having a
will and life cycle savings (1 percent significance level), voluntary pension savings and voting
for European elections (5 percent), and telemarketing (10 percent significance level). Home
owners are more likely to have a will, to join both the new life cycle savings arrangements
and supplementary retirement schemes, to vote and to get rid of telemarketing. Particularly
strong is the magnitude of the marginal effect for the will domain: being home owner
increases the probability of having a will by 35 percentage points. This very strong effect
might in part explain the insignificant role of both income and financial assets in this domain.
The marginal effects for the other domains are smaller, but still in the order of 4-12
percentage points.
The financial situation (gross personal income and household financial assets) does
not seem to play a very significant role in respondents’ behavior regarding non-economic
domains. In pension decisions the financial situation matters a lot. Richer individuals are
more likely to have both voluntary pension savings and life cycle (early retirement) savings.
6. Default choices in the US: evidence from the RAND American Life Panel
The empirical analysis for the Netherlands is based on a questionnaire that was added
to the Dutch DNB Household Survey and fielded in 2006. We have devised a similar module
for the United States, by including the questions in the RAND American Life Panel (ALP).22
This way, we can compare two countries with their own culture and institutional background.
Historically, the US population is used to more freedom of choice in many situations (e.g. in
pension savings), while the Netherlands has a more generous system of social security which
22 We are grateful to Arie Kapteyn and Arthur van Soest for pointing us at this opportunity and their help in entering the formal application procedure. A description of the RAND American Life Panel is available at the website of RAND (http://www.rand.org/labor/roybalfd/american_life.html).
162
impacts labor market decisions and the consequences of unemployment or disability. The
comparison may not only shed light on the impact of culture on decision behavior, but it also
provides information on the robustness of the methodology employed for the Netherlands to
investigate default choices and link these to individual traits.
6.1 Data
The American Life Panel is a joint project between RAND and the University of
Michigan modeled in the spirit of the panel run by CentERdata. Households without an
internet connection are provided with the necessary technology to participate through their
television (a so-called Web TV). They are selected via the University of Michigan’s Survey
of Consumers which interviews a representative sample of the US population. Newly selected
members run through the existing waves of the ALP. All members within the households are
allowed to participate. Participants are interviewed four to six times a year for at most 30
minutes per time. This means that the number of respondents to our module increases with
time. The current sample size equals 1038 individuals with new respondents added as time
evolves and new panel members are being recruited. Contrary to the members of the Dutch
household panel, the ALP participants are paid for their cooperation ($20 for a 30 minutes
survey).
The age of the respondents in our sample ranges from 18 to 87 years (mean age: 50.2
years). Women are slightly in the majority (54.9 percent). As regards education, somewhat
more than 2 out of 10 respondents have a college education, about 6 out of 10 have an
intermediate education level (having some college) and about 2 out of 10 have less education
(until and including high school graduates). High income households are overrepresented as
41.7 percent of the respondents belong to the top quartile for disposable household income;
16.7 percent are in the lowest income quartile, and the other respondents are about evenly
distributed among the second and the third income quartile. Overall, 19.6 percent of
respondents are retired, and 63.0 percent are employed. As for household composition, 63.4
percent of the respondents are married or living with a partner. No information about children
is available.
6.2 Identification of personal traits: evidence from the US
The US ALP survey contains the same 17 statements on personal attitudes and choice
behavior as the Dutch DHS equivalent. Table B1 in appendix B summarizes the responses.
Applying the principal components analysis to the US data delivers results that are mostly
163
similar to the Dutch case. Again five factors have been retained (reported in Table B2 in
appendix B), easily traceable to the ones elicited for the Netherlands. Three out of seventeen
questions are attributed to another factor.23 This illustrates that the factors touch upon
personal characteristics that might be interrelated to some extent. Overall, however, the
resemblance of the findings for the US and the Netherlands seems to confirm the soundness
of the methodology and makes us rather confident on the validity of the information conveyed
by the personal traits stemming from the factor analysis.
Table 5. Personal traits and individual background characteristics in the US _____________________________________________________________________________________ Procrastination Trust Inertia Endorsement Financial literacy ___________ _______ _______ __________ ______________ Male 0.15** -0.42*** -0.11 -0.22*** 0.35*** (2.10) (6.00) (1.57) (3.17) (4.92) Age dummy (36-50) -0.073 -0.14 -0.15 0.080 0.31*** (0.65) (1.24) (1.27) (0.78) (2.78) Age dummy (51-65) -0.32*** -0.26** -0.21* 0.12 0.31*** (3.04) (2.43) (1.90) (1.22) (3.01) Age dummy (>65) -0.60*** -0.16 -0.45*** 0.53*** 0.49*** (3.89) (1.05) (2.87) (3.53) (3.17) Intermediate education 0.088 0.29*** 0.090 -0.15 0.023 (0.80) (2.85) (0.79) (1.37) (0.20) College education 0.13 0.27** 0.28** -0.24* 0.054 (1.05) (2.19) (2.16) (1.88) (0.41) Retired 0.024 0.15 0.16 0.020 -0.073 (0.22) (1.43) (1.44) (0.17) (0.62) Married -0.11 0.040 0.10 -0.006 -0.29*** (1.41) (0.46) (1.26) (0.08) (3.64) Income quartile 2 -0.20* -0.13 -0.19 -0.002 0.064 (1.66) (1.05) (1.49) (0.01) (0.53) Income quartile 3 -0.16 -0.030 -0.056 -0.051 0.094 (1.26) (0.24) (0.46) (0.41) (0.77) Income quartile 4 -0.20 -0.036 -0.009 -0.18 0.13 (1.58) (0.29) (0.08) (1.49) (1.08) Constant 0.31** 0.12 0.083 0.19 -0.37*** (2.22) (0.88) (0.56) (1.41) (2.68) Observations 809 809 809 809 809 R-squared 0.05 0.06 0.03 0.06 0.06 p-value test age = 0 0.00 0.10 0.04 0.00 0.01 p-value test education = 0 0.58 0.02 0.04 0.17 0.91 p-value test income = 0 0.37 0.72 0.35 0.25 0.75 _____________________________________________________________________________________ Note: OLS estimation results; Standard errors are clustered at the household level: absolute value of robust t-statistics in parentheses; * significant at 10%; ** significant at 5%; *** significant at 1%. The dependent variables are the output of the principal component factors analysis. The reference group for age contains respondents less than 36 years, for education respondents with low education and for income respondents in the bottom quartiles.
23 The question ‘If someone tells me to something, I tend to do the opposite’ moves from the endorsement factor to procrastination. The question ‘When I have to buy products requiring specific expertise, I follow the advice of experts’ moves from inertia to trust, replacing the question ‘When making important decisions, I usually take these decisions on my own’ which goes to the literacy factor.
164
As before, we regress each of the five extracted components on the available
background characteristics. Table 5 reports the results. A comparison with the Dutch data is
hampered due to the lower number of observations (809 versus 1422) and the unavailability
of some controls used previously (self-employed dummy, financial assets hold, having
children and being a home owner). Nevertheless, the effect of gender is broadly similar across
both countries. Compared to women, men procrastinate more, have less trust in advice, and
are more confident on their financial literacy. In addition, US men seem to pay less attention
to the opinion of other people than women. Elderly seem to procrastinate less, both in the US
and the Netherlands. The pattern of other age (and to some extent education) coefficients
shows somewhat more differences but these could also be related to the association between
age and education on the one hand and the missing variables in the US specification on the
other hand (e.g. age and education might be related to home ownership and financial assets).
6.3 Choice behavior: evidence from the US
The analysis for the US involves a smaller number of domains with a default option
than for the Netherlands as a result of the exclusion of inapplicable situations like European
elections, stickers on mailbox, and the typical Dutch life cycle savings arrangements for early
retirement. In addition, the automatically renewed subscriptions domain has been dropped due
to the low number of observations. Tables 6A, 6B and 6C report the descriptive statistics for
organ donation, voting participation at the Presidential and local level, having a will,
telemarketing and pension savings. The most striking difference with the Netherlands is that
in the US only 18 percent of respondents stick to the default option of not taking any action to
prevent them from receiving telemarketing contacts, versus 85 percent in the Netherlands.
Moreover, in interpreting the figures for voluntary additional pension savings an important
caveat should be taken into account, as the pension systems differ substantially in the two
countries, thus affecting individual pension savings decisions.
Following the same procedure as before, we first relate the choice behavior in these
situations to the personal traits extracted from the principal component analysis and thereafter
include other individual background characteristics. Tables 7 and 8 report the results. Overall,
it seems somewhat more difficult to adequately describe choice behavior in the US which
might be related to a loss of efficiency due to the smaller sample size and the fact that we
have less information on background characteristics. Nevertheless, procrastination and self-
assessed financial literacy come forward as the most important personal attitude variables.
However, in the US financial literacy appears to be relatively more important whereas in the
165
Table 6A. Organ donation in the US __________________________________________________________________________________ Do you think in general
people ought to be prepared to be an organ donor?
Are you willing to be an organ donor?
Are you an organ donor, i.e. have you signed an affidavit?
_______________________ ______________________ _______________________ Frequency Percentage Frequency Percentage Frequency Percentage __________ __________ __________ __________ __________ __________
Yes 689 66.8 774 72.1 445 59.8 No 116 11.2 119 11.5 254 34.1 Refusal 47 4.5 35 3.4 13 1.8 DK 180 17.4 134 13.0 32 4.3 __________ __________ ________ ________ ________ ________ Total 1032 100.0 1032 100.0 774 100.0 __________________________________________________________________________________ Note: Default option in bold; DK = ‘I do not know’. Table 6B. Voting participation in the US __________________________________________________________________________________ Do you think in general
people ought to vote? Did you vote last time for the Presidential elections?
Did you vote last time for the local elections?
______________________ ______________________ ______________________ Frequency Percentage Frequency Percentage Frequency Percentage __________ __________ __________ __________ __________ __________
Yes 981 95.1 860 87.7 742 75.6 No 23 2.2 106 10.8 221 22.5 Refusal 20 1.9 13 1.3 11 1.1 DK 8 0.8 2 0.2 7 0.7 __________ __________ __________ __________ __________ __________ Total 1032 100.0 981 100.0 981 100.0 __________________________________________________________________________________ Note: Default option in bold; DK = ‘I do not know’ Table 6C. Will, telemarketing, additional retirement savings in the US __________________________________________________________________________________ Do you have a will? Have you registered
yourself in order not to receive telemarketing?
Do you have any other arrangements for your pension apart from Social Security and company pension plans or defined contribution plans?
____________________ ____________________ ______________________________ Frequency Percentage Frequency Percentage Frequency Percentage _________ _________ _________ _________ _________ _________
Yes 509 49.3 823 79.3 464 45.0 No 506 49.0 185 17.8 483 46.8 Refusal 16 1.6 11 1.1 40 3.9 DK 1 0.1 13 1.3 45 4.4 _________ _________ _________ _________ _________ _________ Total 1032 100.0 1032 100.0 1032 100.0 __________________________________________________________________________________ Note: Default option in bold; DK = ‘I do not know’.
166
Table 7. Default choices and personal traits in the US __________________________________________________________________________________ Organ
donation Voting national
Voting local
Will Phone marketing
Additional pension savings
__________ ________ ______ __________ __________ ____________ Procrastination 0.027 0.009 0.030** 0.067*** 0.005 0.081*** (1.23) (0.78) (1.97) (3.34) (0.36) (3.61) Trust 0.001 -0.002 0.028* -0.013 0.002 -0.005 (0.04) (0.22) (1.72) (0.65) (0.15) (0.24) Inertia -0.020 -0.019* 0.001 -0.009 -0.023 -0.062*** (0.92) (1.66) (0.06) (0.48) (1.50) (2.70) Endorsement 0.037* -0.003 -0.027* -0.004 0.013 0.049** (1.69) (0.28) (1.75) (0.20) (0.91) (2.23) Financial literacy -0.025 -0.009 -0.007 -0.031 -0.019 -0.091*** (1.09) (0.73) (0.42) (1.57) (1.42) (3.85) Observations 525 761 761 772 796 579 Pseudo R-squared 0.01 0.01 0.01 0.02 0.01 0.05 p-value test coeff. = 0 0.37 0.58 0.05 0.00 0.53 0.00 __________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, and 0 otherwise.
Netherlands we find a bigger role for procrastination. The most important difference however
is that the endorsement variable that seems to measure social interactions and peer effects
does not play a role in US choice behavior while it was influential in the Netherlands.
7. Discussion
This paper explores individual traits that might explain why default options attract a
disproportionally high number of decision-makers. To the best of our knowledge, it is the first
contribution that relates individual choices in very different situations with a default option to
an extensive set of individual background information including several personal traits and
behavioral attitudes potentially responsible for default choices. Since these behavioral
attitudes and personal traits are not observed directly, we have developed measures based
upon statements on choices that respondents have made or would make in several real-life
situations. The motivation is that people possess intrinsic traits that characterize their
personality and basically guide their behavior in many situations.
We study how individuals decide upon pension savings (both for old age and early
retirement), organ donation, having a will, voting participation, and how they deal with the
cancellation of subscriptions and no-consent choices towards receiving marketing by mail or
phone. These very heterogeneous choice situations all have a default option; i.e. the option
that results if no action is taken. Our analysis is explorative and the measurement of the
relevant personal traits may benefit from an extensive testing of the information contained in
167
simple statements on actual choice behavior. More research on these topics in different
settings with other datasets may shed light on the robustness of our results. Nevertheless, the
fact that the estimation procedure delivers plausible results on the identification of personal
traits and their relation to individual decision-making justifies some confidence in the benefits
of this approach. Especially since this is true for two different counties, the Netherlands and
the US, with their own culture, traditions and institutions.
Table 8. Default choices and personal traits plus background characteristics in the US __________________________________________________________________________________________ Organ
donation Voting national
Voting local
Will Phone marketing
Additional pension savings
__________ _________ _________ _________ __________ ____________ Male 0.15*** -0.017 -0.035 0.025 0.093*** -0.037 (3.50) (1.02) (1.18) (0.62) (3.48) (0.79) Age dummy (36-50) -0.018 -0.062*** -0.098** -0.20*** -0.047 -0.15** (0.26) (3.52) (2.48) (3.05) (1.29) (2.18) Age dummy (51-65) 0.017 -0.10*** -0.15*** -0.37*** -0.049 -0.23*** (0.26) (5.04) (3.76) (5.88) (1.39) (3.32) Age dummy (>65) -0.028 -0.11*** -0.21*** -0.53*** -0.077 -0.27* (0.29) (4.40) (4.44) (7.77) (1.59) (1.65) Intermediate education -0.17** -0.12*** -0.083* -0.031 0.007 -0.29*** (2.45) (4.61) (1.86) (0.52) (0.20) (4.20) College education -0.19*** -0.076*** -0.13*** -0.12* -0.050 -0.33*** (2.70) (3.97) (2.97) (1.72) (1.13) (4.47) Retired 0.017 -0.000 -0.049 -0.016 -0.079** 0.036 (0.25) (0.01) (1.00) (0.25) (2.11) (0.43) Married -0.11** -0.055*** -0.068* -0.082* -0.089*** 0.078 (2.24) (2.80) (1.93) (1.77) (2.80) (1.30) Income quartile 2 -0.043 -0.020 0.029 -0.15** -0.065* -0.092 (0.55) (0.88) (0.59) (2.18) (1.74) (1.09) Income quartile 3 0.094 -0.042* -0.003 -0.18** -0.091** -0.32*** (1.19) (1.95) (0.07) (2.57) (2.46) (4.28) Income quartile 4 0.14* -0.044* 0.016 -0.27*** -0.14*** -0.44*** (1.72) (1.84) (0.33) (4.00) (3.26) (5.40) Procrastination 0.020 -0.004 0.015 0.036* -0.012 0.083*** (0.86) (0.51) (0.99) (1.66) (0.89) (3.24) Trust 0.017 -0.003 0.017 -0.024 0.008 0.011 (0.75) (0.38) (1.10) (1.12) (0.55) (0.47) Inertia -0.027 -0.010 0.008 -0.002 -0.015 -0.046* (1.15) (1.15) (0.48) (0.10) (1.04) (1.83) Endorsement 0.041* -0.006 -0.022 0.001 0.011 0.010 (1.81) (0.74) (1.38) (0.05) (0.81) (0.40) Financial literacy -0.049** -0.001 0.002 -0.035* -0.023* -0.083*** (2.08) (0.10) (0.13) (1.68) (1.78) (3.29) Observations 524 760 760 771 795 579 Pseudo R-squared 0.05 0.19 0.08 0.15 0.09 0.20 p-value test age = 0 0.87 0.00 0.00 0.00 0.37 0.01 p-value test education = 0 0.02 0.00 0.01 0.13 0.25 0.00 p-value test income = 0 0.04 0.20 0.89 0.00 0.01 0.00 __________________________________________________________________________________________ Note: Marginal effects from probit estimates; Standard errors are clustered at the household level: absolute value of robust z-statistics in parentheses; * significant at 10%, ** significant at 5%, *** significant at 1%; For each domain (column), the dependent variable takes value 1 if respondents report to stick to the default option, 0 otherwise; The reference group for age contains respondents less than 36 years, for education respondents with low education and for income respondents in the bottom quartiles.
168
Our descriptive statistics corroborate the stylized fact that default options are a major
attractor in many choice situations. The empirical analysis shows that a large part of the
heterogeneity in individual choice behavior can be explained by objective personal
characteristics and circumstances such as age, education or the financial position.
Nevertheless, procrastination and financial illiteracy prove to be the most important
determinants of default choices in the Netherlands as well as in the US.24 Choices are deferred
because people have an inherent tendency to do so or because of the complicated nature of
choice problems. Moreover, the empirical evidence for the Netherlands suggests that the
extent to which individuals are sensitive to the opinion of others (e.g. through social norms or
peer effects) matters in explaining deviations from the default option.
The latter result raises new questions. Are peer effects and social norms in the
Netherlands indeed more important than in the US? And if so, why is that the case and what
are the implications for policy? One explanation could be that in the Netherlands, a relatively
small densely populated country, the society is more homogeneous than in the US where large
differences with respect to income, education, and racial composition of its population can be
found. If social interactions are relevant for individual decisions, this suggests that publicity
campaigns might play an important role as well as the behavior of policymakers and public
persons in so far this information and these people, respectively, influence social norms.
In the US there seems to be a larger role for financial illiteracy; whereas in the
Netherlands procrastination appears relatively more important. While we can only speculate
about the reasons for this divergence, it could be that whereas in the US private schools are
not uncommon, the Dutch system of public schooling historically might have been more
focused on the provision of a common education contributing to a less pronounced role of
literacy. At the same time, US citizens are accustomed to more freedom of choice and
individual responsibility and might therefore be used to act more decisively reducing the
relevance of procrastination. While the cause of these differences is important in itself, its
explanation goes beyond the scope of this paper.
For policy responses however the relative importance of different explanations is very
relevant. Our estimation results suggest that in the US the provision of information, educating
the public and simplifying choice situations might be the most effective policy instruments to
affect decisions without changing the default. While also relevant for the Netherlands, it
24 We have also experimented with the inclusion of interaction effects as one could argue for example that the impact of financial illiteracy is stronger for individuals who are more likely to procrastinate anyway. We did not find empirical evidence for the importance of such interaction effects though.
169
might be equally effective to deal with the consequences of procrastination, for example by
increasing awareness. Recent experiments on raising the number of organ donor registrations
by presenting a registration form to anyone who enters the town hall to renew a passport
might be viewed in this perspective. Thereby, the existence and the urgency of the donor
register are brought under the attention of citizens every five years instead of once upon
turning eighteen years old. The introduction of a legal obligation for pension funds to send
their participants a pension letter with an overview of pension rights in the form of some
simple scenarios is another example of increasing awareness of the Dutch public.
The pension domain is an important example where both financial illiteracy and
procrastination are relevant for household financial behavior in the US as well as in the
Netherlands. This stresses the need of easily accessible and comprehensible information about
pension products and a constant need to induce people to think about these decisions.
Alternatively, this could motivate a design of the retirement savings system as to prevent
procrastinators from poverty in old age. Moreover, the finding for the Netherlands that
procrastination matters for early retirement savings suggests that the recent redesign of early
retirement institutions in the Netherlands from collective to individual arrangements might
turn out to be very effective in increasing the average retirement age, illustrating the relevance
of default behavior for public policy.
Regarding voluntary and early retirement savings in the Netherlands, we also find a
role for trust and peer effects, respectively. One interpretation is that employees assess the
compulsory nature of employer pension savings as a well-thought advice with no need for
voluntary additional savings. For the life cycle savings accounts, the association with being a
successor of former early retirement arrangements – as advertised by trade unions and
employers - might have induced some employees to deviate from the no-action default of
nonparticipation. These effects come on top of the usual results that employees with better
income and wealth positions (who can afford to save for early retirement or additional old age
provisions) are also more likely to deviate from the default of non-saving, and that higher
educated individuals ceteris paribus show less interest in early retirement as they might have
more challenging or less physically demanding jobs. The fact that compulsory pension
savings (or the publicity around new pension products or arrangements) might have an impact
on savings outcomes because interpreted as an endorsement assigns a lot of responsibility to
governments, trade unions and pension funds in developing ‘optimal’ designs and explaining
their consequences.
170
8. Concluding remarks
This paper contributes to the literature on explaining the relevance of the default
option in decision-making. The adopted approach is innovative as it considers choice
behavior in very different situations and relates these choices to a large set of personal traits
and behavioral attitudes. While more research is needed to validate the results, we believe that
it is worthwhile to further pursue this approach.
The results suggest that overall procrastination and financial illiteracy are important
determinants of decisions. Nevertheless the relative importance depends upon the specific
situation and differs across counties. The implication is that there is no straightforward advice
for the use of default options in public policy. The use of defaults should for instance depend
upon the heterogeneity or homogeneity in the preferences of decision-makers (Beshears,
Choi, Laibson and Madrian, 2007). In addition, the results underline the importance of
simplifying decision processes, where possible, and of informing and educating the public to
increase awareness and help them in making decisions.
171
References Abadie, A., and S. Gay, 2006, The impact of presumed consent legislation on cadaveric organ donation: A cross-country study, Journal of Health Economics, 25, 599-620. Agnew, J., and L. Szykman, 2005, Asset allocation and information overload: The influence of information display, asset choice and investor experience, Journal of Behavioral Finance, 6, 57-70. Agnew, J., L. Szykman, S. Utkus, and J. Young, 2007, Literacy, trust and 401(k) savings behavior, Center for Retirement Research Working Paper, 2007-10. Akerlof, G., 1991, Procrastination and obedience, American Economic Review, 81, 1-19. Alessie, R., S. Hochguertel, and A. van Soest, 2006, Non-take-up of tax-favored savings plans: Evidence from Dutch employees, Journal of Economic Psychology, 27, 483-501. Bernheim, D., 1994, A theory of conformity, Journal of Political Economy, 102, 841-877. Beshears, J., J. Choi, D. Laibson, and B. Madrian, 2007, The importance of default options for retirement saving outcomes: Evidence from the United States, mimeo, Harvard University. Brown, J., Z. Ivkovic, P. Smith, and S. Weisbenner, 2008, Neighbors matter: Causal community effects and stock market participation, Journal of Finance, 63, 1509-1531. Bütler, M., and F. Teppa, 2007, The choice between an annuity and a lump-sum: Results from Swiss pension funds, Journal of Public Economics, 91, 1944-1966. Chang, L., and J. Krosnick, 2003, Comparing oral interviewing with self-administered computerized questionnaires: an experiment, mimeo, Ohio State University. Choi, J., D. Laibson, B. Madrian, and A. Metrick, 2004, For better or for worse: Default effects and 401(k) savings behavior. In: Z. Bodie, J. Shoven, and D. Wise (eds.), Perspectives on the Economics of Aging, University of Chicago Press, Chicago, 81-126. Cronqvist, H., and R. Thaler, 2004, Design choices in privatized social-security systems: Learning from the Swedish experience, American Economic Review, 94, 424-428. Dhar, R., 1997, Consumer preference for a no-choice option, Journal of Consumer Research, 24, 215-231. Duflo, E., and E. Saez, 2002, Participation and investment decisions in a retirement plan: The influence of colleagues’ choices, Journal of Public Economics, 85, 121-148. Duflo, E., and E. Saez, 2003, The role of information and social interactions in retirement plan decisions: Evidence from a randomized experiment, Quarterly Journal of Economics, 118, 815-842. Guiso, L., P. Sapienza, and L. Zingales, 2008, Trusting the stock market, Journal of Finance, forthcoming.
172
Huberman, G., and W. Jiang, 2006, Offering versus choice in 401(k) plans: Equity exposure and number of funds, Journal of Finance, 61, 763-801. Hong, H., J. Kubik, and J. Stein, 2004, Social interaction and stock-market participation, Journal of Finance, 59, 137–163. Iyengar, S. and E. Kamenica, 2008, Choice proliferation, simplicity seeking, and asset allocation, mimeo, University of Chicago. Iyengar, S., and M. Lepper, 2000, When choice is demotivating: Can one desire too much of a good thing?, Journal of Personality and Social Psychology, 79, 995-1006. Johnson, E., S. Bellman, and G. Lohse, 2002, Defaults, framing, and privacy: Why opting in-opting out?, Marketing Letters, 13, 5-15. Johnson, E., and D. Goldstein, 2003, Do defaults save lives?, Science, 302, 1338-1339. Johnson, E., J. Hershey, J. Meszaros, and H. Kunreuther, 1993, Framing, probability distortions, and insurance decisions, Journal of Risk and Uncertainty, 7, 35-51. Kahneman, D., and A. Tversky, 1979, Prospect theory: An analysis of decision making under risk, Econometrica, 47, 263-291. Kahneman, D., and A. Tversky, 1982, The psychology of preferences, Scientific American 246, 160-173. Kapteyn, A., and F. Teppa, 2002, Subjective measures of risk aversion and portfolio choice, mimeo, RAND. Laibson, D., 1997, Golden eggs and hyperbolic discounting, Quarterly Journal of Economics, 112, 443-477. Landman, J., 1987, Regret and elation following action and inaction, Personality and Social Psychology Bulletin, 13, 524-536. Madrian, B., and D. Shea, 2001, The power of suggestion: Inertia in 401(k) participation and savings behavior, Quarterly Journal of Economics, 116, 1149-1187. McKenzie, C., M. Liersch, and S. Finkelstein, 2006, Recommendations implicit in policy defaults, Psychological Science, 17, 414-420. O’Donoghue, T., and M. Rabin, 1999a, Doing it now or later, American Economic Review, 89, 103-124. O’Donoghue, T., and M. Rabin, 1999b, Procrastination in preparing for retirement. In: H. Aaron (ed.), Behavioral Dimensions of Retirement Economics, Brookings Institution Press, Washington DC, 125-156.
173
O’Donoghue, T., and M. Rabin, 2001, Choice and procrastination, Quarterly Journal of Economics, 116, 121-160. Phelps, E., and R. Pollak, 1968, On second-best national savings and game-equilibrium growth, Review of Economic Studies, 35, 185-199. Pollak, R., 1968, Consistent planning, Review of Economic Studies, 35, 201-208. Ritov, I., and J. Baron, 1992, Status-quo and omission biases, Journal of Risk and Uncertainty, 5, 49-61. Samuelson, W., and R. Zeckhauser, 1988, Status quo bias in decision making, Journal of Risk and Uncertainty, 1, 7-59. Strotz, R., 1956, Myopia and inconsistency in dynamic utility maximization, Review of Economic Studies, 23, 165-180. Tversky, A., and E. Shafir, 1992, Choice under conflict: The dynamics of deferred decisions, Psychological Science, 6, 358-361. Van Els, P., M. van Rooij, and M. Schuit, 2007, Why mandatory retirement saving? In: O. Steenbeek and S. van der Lecq (eds.), Costs and Benefits of Collective Pension Systems, Springer, Berlin, 159-186. Van Rooij, M., A. Lusardi, and R. Alessie, 2007, Financial literacy and stock market participation, NBER Working Paper, 13565. Voogt, R., and W. Saris, 2003, To participate or not to participate: The link between survey participation, electoral participation, and political interest, Political Analysis, 11, 164-179.
174
App
endi
x A
. Sev
ente
en s
tate
men
ts a
bout
atti
tude
s an
d ch
oic
e be
havi
or in
the
Net
herla
nds:
dat
a an
d pr
inci
pal c
ompo
nent
anal
ysis
Tab
le A
1. S
tate
men
ts a
bout
per
sona
l atti
tude
s an
d ch
oice be
havi
or in
the
Net
herla
nds:
Wor
ding
and
sum
mar
y of
res
ponse
s P
erce
nta
ges
of t
ota
l nu
mb
er o
f re
spon
dent
s (N
=16
48)
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
P
lea
se in
dica
te o
n a
sca
le f
rom
1 t
o 7
to w
hat
exte
nt y
ou a
gre
e w
ith e
ach
of
the
follo
win
g st
ate
men
ts (
1 m
eans
‘tot
ally
dis
agr
ee’ a
nd 7
mea
ns ‘t
ota
lly a
gree
’)?
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
_
1
2 3
4 5
6 7
D
K
M
ean 2)
__
_ __
_ __
_ __
_ __
_ __
_ __
_
___
__
____
W
hen
ma
king
imp
orta
nt d
ecis
ions
(e.
g. b
uyin
g a
car
or
inve
stin
g m
oney
)
Q
1 -
I usu
ally
ta
lk w
ith o
ther
peo
ple
abo
ut it
3,
1 8,
6 8,
0 13
,8
25,1
27
,4
11,7
2,1
4,
8 (1
,6)
Q
2 -
I usu
ally
ta
ke t
hese
dec
isio
ns o
n m
y ow
n 19
,8
24,9
16
,3
9,7
9,7
11,4
5,
5
2,4
3,
2 (1
,9)
Q
3 -
I usu
ally
lea
ve it
to
som
eone
els
e 37
,6
29,3
13
,8
8,0
5,0
2,7
1,4
1,
9
2,3
(1,4
) W
hen
I ha
ve t
o bu
y pr
oduc
ts r
equi
ring
sp
ecifi
c ex
per
tise
(e.g
. a
fina
ncia
l or
a te
chno
logi
cal p
rodu
ct)
Q
4 -
I fol
low
the
adv
ice
of e
xper
ts
0,9
1,2
5,3
14,8
34
,4
31,8
9,
2 2,
4
5,2
(1,1
) Q
5 -
I ta
lk a
bout
it w
ith f
am
ily o
r fr
iend
s 3,
0 7,
2 7,
9 14
,7
29,5
24
,6
10,6
2,
4
4,8
(1,5
) Q
6 -
I of
ten
rely
on
wha
t pe
ople
sa
y 8,
9 23
,3
24,2
25
,7
12,3
3,
3 0,
4 1,
6
3,2
(1,3
) Q
7 -
If so
meo
ne t
ells
me
to d
o so
met
hing
, I
tend
to
do t
he o
ppos
ite
10,9
24
,3
17,7
25
,0
12,3
5,
1 2,
3 2,
2
3,3
(1,5
) Q
8 -
I us
ually
do
wha
t ot
her
peop
le t
ell m
e to
do
13,8
26
,7
22,2
21
,4
10,7
2,
7 0,
4 1,
8
3,0
(1,4
) Q
9 -
I ha
ve t
roub
les
to s
ay
no t
o pe
ople
6,
4 14
,8
12,1
13
,5
25,9
18
,5
7,0
1,6
4,
2 (1
,7)
Q1
0 -
I do
cho
res
right
aw
ay
4,7
14,4
23
,4
21,2
17
,8
12,7
4,
5 1,
1
3,9
(1,5
) Q
11
- I
tend
to
ma
ke p
rom
ises
tha
t I
cann
ot k
eep
29,4
38
,8
13,6
8,
1 5,
9 1,
7 0,
6 1,
6
2,3
(1,3
) Q
12
- W
hen
I pr
omis
e to
do
som
ethi
ng,
I us
ually
do
tha
t la
ter
tha
n I
shou
ld
18,8
34
,4
16,6
11
,0
12,3
3,
6 1,
6 1,
5
2,8
(1,5
) Q
13
- C
hang
es a
re s
cary
12
,7
26,9
20
,4
20,7
13
,1
3,3
1,0
1,6
3,
1 (1
,4)
Q1
4 -
Cha
nges
are
oft
en n
ot a
n im
prov
emen
t 4,
3 13
,8
14,5
30
,4
16,6
11
,8
6,0
2,5
4,
0 (1
,5)
Q1
5 -
I w
ould
des
crib
e m
ysel
f a
s a
ca
refu
l per
son
1,6
5,6
11,9
20
,9
27,9
23
,7
6,6
1,5
4,
7 (1
,4)
Q1
6 -
Whe
n th
ere
is p
ossi
ble
da
nger
, I
take
ma
ny p
reca
utio
ns
0,4
4,
1 8,
0 17
,4
31,9
25
,5
10,1
2,
4
5,0
(1,3
) Q
17
- H
ow w
ould
you
ass
ess
your
fin
anc
ial s
kills
1)
2,4
6,7
12,3
25
,4
31,8
16
,9
1,5
2,7
4,
4 (1
,3)
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
1)
For
this
que
stio
n r
espo
nden
ts a
re a
sked
to in
dica
te th
eir
skill
s on
a 7
-poi
nts
sca
le, w
her
e 1
mea
ns
‘ver
y ba
d’ a
nd
7 m
ean
s ‘v
ery
good
’. 2)
Mea
n re
fers
to
the
ave
rage
of t
he
seve
n r
espo
nse
ca
teg
orie
s fr
om 1
to
7 (s
tan
dard
dev
iatio
n in
par
enth
eses
).
Not
e: D
K =
‘I d
o n
ot k
now
’; D
K’s
an
d re
spon
se c
ate
gori
es 1
-7 d
o n
ot s
um u
p to
100
% d
ue to
ref
usa
ls.
175
Tab
le A
2. S
tate
men
ts a
nd t
he h
ighe
st f
acto
r lo
adin
gs fro
m a
prin
cipa
l co
mpo
nent
ana
lysi
s fo
r th
e N
ethe
rland
s __
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
S
tate
men
ts
Fa
ctor
s
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
__
____
____
____
____
____
____
____
____
____
____
____
___
P
rocr
astin
atio
n T
rust
In
ertia
End
orse
men
t F
ina
ncia
l lit
era
cy
__
____
____
__
____
_ __
___
____
____
__
____
____
I
do c
hore
s rig
ht a
wa
y -0
,64
I te
nd t
o m
ake
pro
mis
es t
hat
I ca
nnot
kee
p 0,
69
Whe
n I
prom
ise
to d
o so
met
hing
, I
usua
lly d
o th
at
late
r th
an
I sh
ould
0,
80
Whe
n m
aki
ng im
por
tant
dec
isio
ns,
I us
ually
ta
lk w
ith o
ther
pe
ople
abo
ut it
0,84
Whe
n m
aki
ng im
por
tant
dec
isio
ns,
I us
ually
ta
ke t
hese
dec
isio
ns
on m
y ow
n
-0,5
5
Whe
n I
have
to
buy
prod
ucts
req
uiri
ng s
pec
ific
exp
ertis
e, I
ta
lk a
bout
it w
ith f
am
ily o
r fr
iend
s
0,77
Whe
n I
have
to
buy
prod
ucts
req
uiri
ng s
pec
ific
exp
ertis
e, I
fol
low
the
adv
ice
of e
xper
ts
0,36
C
hang
es a
re s
cary
0,
55
Cha
nges
are
oft
en n
ot a
n im
prov
emen
t
0,
64
I w
ould
des
crib
e m
ysel
f a
s a
care
ful p
erso
n
0,
73
Whe
n th
ere
is p
ossi
ble
da
nger
, I
take
ma
ny p
reca
utio
ns
0,62
I
ofte
n re
ly o
n w
hat
peop
le s
ay
0,
69
If
som
eone
tel
ls m
e to
do
som
ethi
ng,
I te
nd t
o do
the
opp
osite
0,
48
-0,4
8
I us
ually
do
wha
t ot
her
peop
le t
ell m
e to
do
0,
76
I
have
tro
uble
s to
sa
y no
to
peo
ple
0,47
Whe
n m
aki
ng im
por
tant
dec
isio
ns,
I us
ually
lea
ve it
to
som
eone
els
e
-0
,69
How
wou
ld y
ou a
sses
s yo
ur f
ina
ncia
l ski
lls
0,73
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
N
ote:
For
ea
ch o
f th
e st
atem
ents
we
repo
rt th
e hi
ghes
t fa
ctor
loa
din
g fr
om a
prin
cipa
l com
pon
ent a
naly
sis
usin
g va
rim
ax
rota
tion;
N =
150
9.
176
App
endi
x B
. Sev
ente
en s
tate
men
ts a
bout
atti
tude
s an
d ch
oice b
ehav
ior
in th
e U
nite
d S
tate
s: d
ata
and
prin
cipa
l com
ponen
t ana
lysi
s
Tab
le B
1. S
tate
men
ts a
bout
per
sona
l atti
tude
s an
d ch
oi
ce b
ehav
ior
in t
he U
S: W
ordi
ng a
nd s
umm
ary
of r
espo
nses
P
erce
nta
ges
of t
ota
l nu
mb
er o
f re
spon
dent
s (N
=10
38)
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
P
lea
se in
dica
te o
n a
sca
le f
rom
1 t
o 7
to w
hat
exte
nt y
ou a
gre
e w
ith e
ach
of
the
follo
win
g st
ate
men
ts (
1 m
eans
‘tot
ally
dis
agr
ee’ a
nd 7
mea
ns ‘t
ota
lly a
gree
’)?
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
_
1
2 3
4 5
6 7
D
K
M
ean 2)
__
_ __
_ __
_ __
_ __
_ __
_ __
_
___
__
____
W
hen
ma
king
imp
orta
nt d
ecis
ions
(e.
g. b
uyin
g a
car
or
inve
stin
g m
oney
)
Q
1 -
I usu
ally
ta
lk w
ith o
ther
peo
ple
abo
ut it
3,
1 5,
9 5,
7 11
,3
21,3
26
,5
25,0
1,
4
5,3
(1,6
)
Q2
- I u
sua
lly t
ake
the
se d
ecis
ions
on
my
own
12,6
15
,6
13,1
15
,2
14,2
17
,0
10,3
2,
0
4,1
(2,0
)
Q3
- I u
sua
lly le
ave
it t
o so
meo
ne e
lse
53,7
25
,7
8,8
5,2
2,4
2,0
0,3
1,9
1,
9 (1
,5)
Whe
n I
have
to
buy
prod
ucts
req
uiri
ng s
pec
ific
exp
ertis
e (e
.g.
a fin
anc
ial o
r a
tech
nolo
gica
l pro
duct
)
Q
4 -
I fol
low
the
adv
ice
of e
xper
ts
1,0
2,5
5,0
16,3
31
,7
28,5
12
,8
2,2
5,
8 (1
,5)
Q
5 -
I ta
lk a
bout
it w
ith f
am
ily o
r fr
iend
s 2,
4 5,
2 7,
9 12
,0
23,4
24
,6
22,8
1,
7
5,2
(1,3
) Q
6 -
I of
ten
rely
on
wha
t pe
ople
sa
y 7,
1 14
,1
19,2
29
,4
21,3
5,
8 1,
6 1,
5
4,4
(1,7
) Q
7 -
If so
meo
ne t
ells
me
to d
o so
met
hing
, I
tend
to
do t
he o
ppos
ite
24,3
31
,3
15,9
19
,0
3,7
1,7
1,3
2,9
1,
9 (1
,6)
Q8
- I
usua
lly d
o w
hat
othe
r pe
ople
tel
l me
to d
o 18
,8
26,0
19
,8
22,5
7,
4 2,
6 0,
8 2,
2
2,3
(1,6
) Q
9 -
I ha
ve t
roub
les
to s
ay
no t
o pe
ople
13
,4
18,8
12
,7
14,1
19
,4
12,9
7,
7 1,
1
4,0
(1,7
) Q
10
- I
do c
hore
s rig
ht a
wa
y 4,
4 9,
2 17
,2
21,9
18
,6
17,4
10
,0
1,4
5,
2 (1
,6)
Q1
1 -
I te
nd t
o m
ake
pro
mis
es t
hat
I ca
nnot
kee
p 47
,1
33,9
9,
3 4,
8 3,
0 1,
5 0,
4 0,
1
1,9
(1,2
) Q
12
- W
hen
I pr
omis
e to
do
som
ethi
ng,
I us
ually
do
tha
t la
ter
tha
n I
shou
ld
25,1
31
,8
15,2
10
,7
9,9
4,1
1,8
1,5
2,
7 (1
,6)
Q1
3 -
Cha
nges
are
sca
ry
8,1
14,3
15
,0
23,8
19
,7
10,2
6,
9 2,
0
3,3
(2,0
) Q
14
- C
hang
es a
re o
ften
not
an
impr
ovem
ent
7,1
14,4
15
,3
26,6
14
,7
8,4
10,0
3,
5
3,1
(1,4
) Q
15
- I
wou
ld d
escr
ibe
mys
elf
as
a c
are
ful p
erso
n 1,
1 1,
9 5,
8 13
,9
25,5
32
,2
18,4
1,
3
4,1
(1,8
) Q
16
- W
hen
ther
e is
pos
sib
le d
ang
er,
I ta
ke m
any
pre
caut
ions
0,
5
1,4
6,4
10,6
21
,8
31,2
26
,5
1,7
5,
6 (1
,3)
Q1
7 -
How
wou
ld y
ou a
sses
s yo
ur f
ina
ncia
l ski
lls 1
) 2,
2 4,
7 10
,5
24,6
31
,9
19,0
5,
2 1,
9
4,7
(1,4
) __
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
__
1) F
or th
is q
uest
ion
res
pond
ents
are
ask
ed to
indi
cate
thei
r sk
ills
on a
7-p
oin
ts s
cale
, wh
ere
1 m
ean
s ‘v
ery
bad’
an
d 7
mea
ns
‘ver
y go
od’.
2) M
ean
refe
rs t
o th
e a
vera
ge o
f th
e se
ven
res
pon
se c
at
egor
ies
from
1 t
o 7
(sta
nda
rd d
evia
tion
in p
aren
thes
es).
N
ote:
DK
= ‘I
do
not
kn
ow’;
DK
’s a
nd
resp
onse
ca
tego
ries
1-7
do
not
sum
up
to 1
00%
due
to r
efus
als
.
177
Tab
le B
2. S
tate
men
ts a
nd t
he h
ighe
st fa
ctor
load
ing
s from
a p
rinci
pal c
om
pone
nt a
naly
sis
for
the
US
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
S
tate
men
ts
Fa
ctor
s
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
__
____
____
____
____
____
____
____
____
____
____
____
___
P
rocr
astin
atio
n T
rust
In
ertia
End
orse
men
t F
ina
ncia
l lit
era
cy
__
____
____
__
____
_ __
___
____
____
__
____
____
I
do c
hore
s rig
ht a
wa
y -0
,63
I te
nd t
o m
ake
pro
mis
es t
hat
I ca
nnot
kee
p 0,
48
Whe
n I
prom
ise
to d
o so
met
hing
, I
usua
lly d
o th
at
late
r th
an
I sh
ould
0,
65
If so
meo
ne t
ells
me
to d
o so
met
hing
, I
tend
to
do t
he o
ppos
ite
0,58
W
hen
ma
king
imp
orta
nt d
ecis
ions
, I
usua
lly t
alk
with
oth
er p
eop
le a
bout
it
0,
80
W
hen
I ha
ve t
o bu
y pr
oduc
ts r
equi
ring
sp
ecifi
c ex
per
tise,
I f
ollo
w t
he a
dvic
e of
exp
erts
0,42
Whe
n I
have
to
buy
prod
ucts
req
uiri
ng s
pec
ific
exp
ertis
e, I
ta
lk a
bout
it w
ith f
am
ily o
r fr
iend
s
0,81
Cha
nges
are
sca
ry
0,63
C
hang
es a
re o
ften
not
an
impr
ovem
ent
0,60
I
wou
ld d
escr
ibe
mys
elf
as
a ca
refu
l per
son
0,59
W
hen
ther
e is
pos
sib
le d
ang
er,
I ta
ke m
any
pre
caut
ions
0,
64
I of
ten
rely
on
wha
t pe
ople
sa
y
0,66
I us
ually
do
wha
t ot
her
peop
le t
ell m
e to
do
0,
74
I
have
tro
uble
s to
sa
y no
to
peo
ple
0,56
Whe
n m
aki
ng im
por
tant
dec
isio
ns,
I us
ually
lea
ve it
to
som
eone
els
e
-0
,63
How
wou
ld y
ou a
sses
s yo
ur f
ina
ncia
l ski
lls
0,65
W
hen
ma
king
imp
orta
nt d
ecis
ions
, I
usua
lly t
ake
the
se d
ecis
ion
s on
my
own
0,61
__
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
____
N
ote:
For
ea
ch o
f th
e st
atem
ents
we
repo
rt th
e hi
ghes
t fa
ctor
loa
din
g fr
om a
prin
cipa
l com
pon
ent a
naly
sis
usin
g va
rim
ax
rota
tion;
N =
941
.
179
SAMENVATTING (SUMMARY IN DUTCH)
181
1. Achtergrond en motivatie
Dit proefschrift bevat vier empirische artikelen over het financiële gedrag van
Nederlandse gezinnen. Inzicht in financiële keuzes en het vermogensbeheer van gezinnen is
om verschillende redenen van belang. Enerzijds bevindt het algemene welvaartsniveau zich
op ongekende hoogte. Dit geeft veel gezinnen de mogelijkheid om geld opzij te legen, te
beleggen en keuzes te maken ten aanzien van verdeling van consumptie, vrije tijd en arbeid
(aantal gewerkte uren en het moment van pensionering) over de levenscyclus. Anderzijds
wordt in toenemende mate van mensen verwacht dat zij individueel verantwoordelijkheid
nemen voor het eigen financiële welzijn. Tegelijkertijd heeft de deregulering van financiële
markten de concurrentie tussen financiële instellingen vergroot en financiële innovaties
gestimuleerd wat onder andere heeft geleid tot een continue stroom van nieuwe financiële
producten. Deze ontwikkelingen dragen bij aan de toegenomen, wereldwijde belangstelling
van economen voor het financiële gedrag van gezinnen (zie de openingsrede voor de
American Finance Association door John Campbell (2006)) en verklaren tevens de
toenemende aandacht in beleidsdiscussies hiervoor aangezien de effectiviteit van budgettair
en monetair beleid in belangrijke mate afhangt van de gedragsreacties van consumenten.
Het levenscyclusmodel voor consumptie- en spaargedrag vormt nog steeds het
belangrijkste startpunt voor een beschrijving van het financiële gedrag van gezinnen. De
meest eenvoudige versies van deze modellen voorspellen dat gezinnen tijdens hun werkzame
leven vermogen opbouwen om daarmee hun pensionering te financieren (Modigliani en
Brumberg, 1954; Friedman, 1957). Sindsdien zijn deze modellen op verschillende manieren
uitgebreid om ze realistischer te maken bijvoorbeeld door rekening te houden met
onzekerheid, liquiditeitsbeperkingen en erfenismotieven (zie Browning en Lusardi (1996)
voor een overzicht). De basale, onderliggende veronderstellingen zijn echter onveranderd in
de meeste modellen; consumenten worden beschouwd als rationele agenten die alle relevante
informatie verzamelen en gebruiken om het verwachte nut over hun resterende levensduur te
maximaliseren. Psychologisch onderzoek betwist deze aanname en beargumenteert dat
consumenten vuistregels gebruiken en dat hun gedrag beter kan worden verklaard uit
psychologische concepten zoals verliesaversie, bijziendheid of mentaal boekhouden.
Tegelijkertijd bestaan diverse voorbeelden van financiële missers van gezinnen die de
impliciete aanname uit het levenscyclusmodel schenden dat huishoudens over voldoende
financiële vaardigheden beschikken om optimale beslissingen te nemen. Het verbindende
element van de artikelen verzameld in dit proefschrift is de rol van financiële vaardigheden bij
financiële beslissingen vooral in relatie tot pensioenkeuzes. In het bijzonder besteden de
182
bijdragen aandacht aan pensioenvoorkeuren van werknemers, hun bereidheid om
pensioeninvesteringen zelf ter hand te nemen en hun capaciteiten om dat te doen; het meten
van financiële kennis en vaardigheden en de gevolgen daarvan voor beleggingsgedrag,
vermogensopbouw en pensioenplanning; en tot slot de rol van financiële geletterdheid en
andere determinanten voor beslissingen in keuzesituaties met een standaardoptie dat wil
zeggen de keuze die impliciet of expliciet wordt gemaakt door geen actie te ondernemen.
De bijdragen in dit proefschrift zijn niet normatief, maar doen een poging het gedrag
van gezinnen te beschrijven en te verklaren met behulp van gegevens die worden verkregen
uit specifiek voor dit doel ontworpen internetenquêtes onder een panel van Nederlandse
huishoudens van CentERdata. In het verleden hebben economen zich behoudend opgesteld
ten aanzien van het nut van enquêtes en het informatiegehalte van subjectieve antwoorden,
maar tegenwoordig worden deze wijdverbreid gebruikt en is overtuigend duidelijk geworden
dat zij nuttige informatie opleveren met voorspellende waarde voor het gedrag van gezinnen.1
Bovendien zijn dit type enquêtes onmisbaar voor het verkrijgen van informatie over
heterogene voorkeuren en houdingen van gezinnen die cruciaal zijn om individuele
beslissingen te begrijpen.
2. Onderzoeksresultaten
Het eerste artikel getiteld ‘Risk-return preferences in the pension domain: Are people
able to choose?’ bestudeert pensioenvoorkeuren en beleggingsgedrag van Nederlandse
werknemers. Nederland vormt een interessante casestudy omdat haar pensioensysteem
nauwelijks enige keuzevrijheid biedt, terwijl de laatste drie decennia wereldwijd sprake is van
een verschuiving van risico’s en verantwoordelijkheid voor pensioenbeleggingen van
werkgevers naar werknemers. De Verenigde Staten en het Verenigd Koninkrijk bijvoorbeeld
hebben een sterke verschuiving laten zien naar beschikbare premieregelingen (Defined
Contribution ofwel DC pensioenregelingen) ten koste van regelingen die uitgaan van een
toezegging ten aanzien van de pensioenuitkering (Defined Benefit of DB-regelingen). Nieuwe
internationale boekhoudstandaarden, de aandelenmarktcrisis van 2000-2003 en de structurele
afname van kapitaalmarktrentes hebben een debat aangezwengeld over de houdbaarheid van
het DB-systeem en of ook in Nederland meer beleggingsvrijheid en meer risico naar
werknemers moet worden geschoven.
1 Zie bijvoorbeeld Hurd en McGarry (2002), Manski (2004), Donkers en Van Soest (1999), en Kapteyn en Teppa (2002).
183
De onderzoeksresultaten laten zien dat een grote meerderheid van de Nederlandse
werknemers tegenstander is van veranderingen in de richting van meer individuele
verantwoordelijkheid voor pensioenen. Deze voorkeuren zijn consistent met hun houding ten
aanzien van risico en een zelfinschatting van de eigen financiële vaardigheden. Respondenten
blijken namelijk in hoge mate risicomijdend, in het bijzonder waar het om pensioenen gaat, en
hebben een sterke voorkeur voor een gegarandeerd inkomen na pensionering. Daarnaast
beschouwt de gemiddelde respondent zichzelf als financieel ondeskundig en is niet bereid om
de zeggenschap over pensioenbesparingen uit te oefenen zelfs als hem de mogelijkheid wordt
geboden zijn financiële expertise te vergroten. Uit een experiment volgt dat respondenten die,
in een denkbeeldig DC-systeem, in eerste instantie kiezen voor een relatief veilige
beleggingsportefeuille veelal switchen naar een meer risicovolle beleggingsstrategie conform
de keuze van de gemiddelde respondent als zij worden geconfronteerd met de gevolgen van
hun keuze voor de kansverdeling van toekomstige pensioenuitkeringen. Dit suggereert dat de
financiële vaardigheden ontoereikend zijn om controle uit te oefenen over hun eigen
pensioenvermogen. Terwijl het zeer wel mogelijk is dat deze uitkomsten deels worden
veroorzaakt door een gebrek aan ervaring met het uitoefenen van invloed op de opbouw van
het pensioenvermogen in het verleden, roepen zij tegelijkertijd vragen op over het algemene
niveau van financiële geletterdheid van de Nederlandse bevolking.
Het tweede artikel getiteld ‘Financial literacy and stock market participation’, richt
zich op het meten van financiële kennis en cognitieve vaardigheden. Niet alleen
pensioenbeslissingen, maar vele financiële keuzes zijn ingewikkelder geworden door
financiële innovaties en het stijgende aanbod van financiële producten zoals op de markt voor
leningen. Terwijl financiële vaardigheden een noodzakelijke voorwaarde zijn om in te spelen
op de toename in individuele verantwoordelijkheid voor het eigen financiële welzijn, staat pas
relatief kort de vraag centraal of consumenten wel in staat zijn zich door deze nieuwe
financiële omgeving te bewegen. Wij hebben een uitgebreide vragenlijst ontworpen om
inzicht te krijgen in basale financiële vaardigheden gerelateerd aan eenvoudige
rekenvaardigheden, de werking van inflatie en rentevoeten en meer gevorderde onderwerpen
gerelateerd aan financiële instrumenten (aandelen, obligaties en beleggingsfondsen). Ons
werk vormt daarbij een stap voorwaarts ten opzichte van eerdere studies door meer verfijnde
maatstaven voor financiële geletterdheid te beschouwen.
Onze gegevens laten zien dat de meerderheid van de respondenten over een zekere
basale, financiële kennis beschikt en enig inzicht heeft in de werking van samengestelde rente,
inflatie en de tijdswaarde van geld. Vaak reikt het kennisniveau echter nauwelijks verder:
184
velen zijn niet goed op de hoogte van het verschil tussen obligaties en aandelen, de relatie
tussen obligatiekoersen en rentevoeten en de voordelen van risicodiversificatie. Tevens
verschaft ons onderzoek informatie over de methodologie van het meten van financiële
kennis: wij tonen aan dat er veel ruis zit in de antwoorden van respondenten. Het blijkt dat de
formulering van de vragen van groot belang is voor het meten van financiële kennis en dat
kleine veranderingen in de woordvolgorde grote consequenties kunnen hebben voor de
beantwoording van de vragen. Deze gevoeligheid voor de formulering van enquêtevragen laat
zien dat het van groot belang is om dergelijke vragen om financiële kennis te meten eerst te
testen en te valideren in een pilot-versie van de enquête, maar vormt tevens een aanvullende
illustratie voor een beperkte financiële kennis.
Wij illustreren het belang van financiële geletterdheid door na te gaan of personen met
een grotere financiële kennis meer kans hebben om in aandelen te beleggen. Hierbij sluiten
wij aan bij de literatuur die de zogenoemde ‘stockholding puzzle’ probeert te verklaren
(Haliassos en Bertaut, 1995). Standaardmodellen gebaseerd op de maximalisatie van het
verwachte nut geven aan dat het voor vrijwel iedereen aantrekkelijk is om op zijn minst een
klein deel, van het vermogen in aandelen te beleggen. De puzzel is dat in de praktijk in vele
landen een grote meerderheid zich niet op de aandelenmarkt begeeft (Guiso, Haliassos en
Jappelli, 2002). In de literatuur is een zekere consensus ontstaan dat een belangrijke rol in de
verklaring is weggelegd voor de kosten van het verzamelen en verwerken van informatie
inclusief bijvoorbeeld de kosten die gemoeid zijn met het uitzoeken hoe men kan beleggen, en
het monitoren van adviseurs en beleggingsuitkomsten. Tegelijkertijd kan dit niet afdoende
verklaren waarom meer vermogende huishoudens niet vaker beleggen in aandelen. Andere
onderzoeken wijzen op het belang van vertrouwen en sociale interacties. Een hoger
kennisniveau verlaagt de informatiekosten en de relevantie van de verschillende drempels
voor aandelenmarktparticipatie. Onze empirische resultaten geven ondersteuning aan deze
visie. Vanwege de mogelijkheid van omgekeerde causaliteit, dat wil zeggen dat respondenten
financiële kennis opdoen door hun activiteiten in de aandelenmarkt, maken wij gebruik van de
variatie in de mate waarin respondenten hebben blootgestaan aan economische scholing, een
maatstaf voor de aanwezige financiële kennis in een periode van het leven waarin de kans dat
mensen al beleggen in de aandelenmarkt uitermate gering is.
Hert derde artikel getiteld ‘Financial literacy, retirement planning, and household
wealth’, richt zich op de gevolgen van financiële geletterdheid voor de netto
vermogenspositie van huishoudens. Daarmee onderzoekt het de relevantie van financiële
kennis voor huishoudgedrag en financiële uitkomsten vanuit een breder perspectief. Er zijn
185
veel voorbeelden bekend van financiële missers door huishoudens en Amerikaans onderzoek
rapporteert enig empirisch bewijs, hoewel niet onbetwist, dat financiële educatie leidt tot meer
besparingen (zie Lusardi (2004) voor een overzicht). Deze studies richten zich echter niet
specifiek op de vraag of het effect van financiële educatie op spaargedrag loopt via een
toename in financiële kennis of via andere kanalen. De gevonden effecten op besparingen
kunnen namelijk – op zijn minst ten dele – ook samenhangen met het beschikbaar stellen van
informatie, het aanbieden van middelen om je te committeren aan besparingen, sociale
interacties of het gevolg zijn van zelfselectie van respondenten in het geval van het bijwonen
van financiële seminars.
Onze schattingsresultaten laten een statistisch en economisch significant effect zien
van financiële kennis op het netto vermogen. Dit is een belangrijk resultaat uit het oogpunt
van overheidsbeleid met het oog op de vrees dat huishoudens wellicht onvoldoende sparen
voor hun pensioen. De onderzoeksresultaten laten inderdaad zien dat financiële kennis een
stimulans vormt om te plannen voor het pensioen. Deze planning gaat gepaard met het
verzamelen en verwerken van informatie over (toekomstige) inkomsten en uitgaven en het
doen van de daarvoor noodzakelijke berekeningen. Dit proces voorziet planners van
informatie over benodigde besparingen, terwijl de gerelateerde activiteiten hen kunnen helpen
eventuele zelfbeheersingproblemen te overwinnen en hun vermogen te vergroten (Ameriks,
Caplin en Leahy, 2003; Lusardi en Mitchell, 2007). Tegelijkertijd draagt het feit dat financiële
kennis de participatie op de aandelenmarkt stimuleert mogelijk ook bij aan een hogere
vermogensopbouw. Toetreding tot de aandelenmarkten vergroot immers de mogelijkheden
voor huishoudens om hun vermogen te spreiden en te profiteren van de risicopremie op
aandelenbeleggingen wat bij kan dragen aan een efficiënter beheer van de
vermogensportefeuille. Beide genoemde kanalen vormen potentiële verklaringen die kunnen
bijdragen aan het positieve effect van financiële kennis op het netto gezinsvermogen.
Tenslotte laten de schattingsresultaten zien dat degenen die relatief veel vertrouwen hebben in
hun eigen financiële vaardigheden ook een grotere kans hebben aan pensioenplanning te doen.
Blijkbaar is de mate waarin mensen zich op hun gemak voelen met hun financiële
vaardigheden een apart element dat mensen al dan niet weghoudt van informatie-intensieve
beslissingen. Dit suggereert dat in aanvulling op financiële educatie inspanningen om
keuzeproblemen op een heldere en begrijpelijke manier te presenteren effectief kunnen zijn
bij de ondersteuning van huishoudens bij het nemen van complexe beslissingen.
Het vierde artikel getiteld ‘Choice or no choice: What explains the attractiveness of
default options?’ onderzoekt het effect van financiële geletterdheid op pensioenbeslissingen
186
vanuit een ander perspectief. In de literatuur is uitgebreid gedocumenteerd dat de
standaardoptie in beslissingsproblemen door een disproportioneel groot aantal beslissers
wordt gekozen. Vanuit theoretisch oogpunt zou de manier waarop keuzeproblemen worden
aangeboden - en in het bijzonder wat de standaardkeuzemogelijkheid is - er niet toe moeten
doen zolang de voorkeuren van de beslisser duidelijk zijn en de kosten van het kiezen uit de
alternatieven verwaarloosbaar klein zijn. Van oudsher is de gedachte altijd geweest dat een
vergroting van het aantal keuzemogelijkheden leidt tot een hoger welvaartsniveau omdat
beslissers ervoor kunnen kiezen nieuwe alternatieven buiten beschouwing te laten. Inzichten
uit psychologisch onderzoek laten echter zien dat informatie overbelasting en keuze stress een
belangrijke invloed uitoefenen op uitkomsten van beslissingsprocessen en een keuze voor de
standaardoptie meer waarschijnlijk maken. Er zijn echter ook andere mogelijke verklaringen
waarom standaard keuzeopties vaker worden gekozen zoals laksheid, een voorkeur voor de
status quo, de interpretatie van standaardopties als een impliciet advies en uitstelgedrag.
Zover wij weten bestaat er geen studie die het relatieve belang van de verschillende
verklaringen empirisch heeft onderzocht. De achterliggende veronderstelling in onze studie is
dat degenen die beslissingen in de ene situatie uitstellen bijvoorbeeld vanwege het complexe
karakter ervan, dit type gedrag mogelijk ook in andere situaties zullen vertonen. Wij
beschouwen een aantal zeer verschillende keuzesituaties met verschillende karakteristieken en
onderzoeken of er een dominante factor is die de keuze voor de standaardoptie verklaart. Op
basis hiervan willen wij lessen trekken voor belangrijke financiële keuzes en in het bijzonder
pensioenbeslissingen.
Onze bevindingen voor Nederlandse respondenten laten zien dat uitstelgedrag en
financiële geletterdheid belangrijke determinanten zijn van het keuzegedrag in situaties met
eens standaardoptie. De situaties die wij beschouwen bevatten belangrijke financiële
beslissingen zoals sparen voor de oude dag of voor vroegpensioen en het afsluiten van een
testament. Daarnaast bestuderen wij beslissingen ten aanzien van de registratie als
orgaandonor, het uitbrengen van je stem tijdens verkiezingen, het opzeggen van
abonnementen, en het aantekenen van bezwaar tegen het ontvangen van marketing via de post
of telefoon. Op basis van de regressieanalyses lijkt een belangrijke rol weggelegd voor sociale
interacties en sociale normen bij de verklaring van afwijkingen van de standaard keuzeoptie
vanwege de invloed van de mening van anderen op keuzegedrag.
Dit artikel bevat ook een vergelijkende analyse voor de Verenigde Staten, aangezien
wij de kans hadden een verkorte versie van de vragenlijst voor te leggen aan de leden van het
RAND American Life Panel. Ook in de VS komen uitstelgedrag en financieel analfabetisme
187
naar voren als belangrijke verklaringen voor de aantrekkelijkheid van de standaardoptie, maar
wij vinden geen rol voor sociale normen en interacties vergelijkbaar met die voor de
Nederlandse situatie. Bovendien lijkt in de VS financiële ongeletterdheid relatief gezien
belangrijker, terwijl in Nederland uitstelgedrag een meer dominante rol vervult. Terwijl wij
alleen kunnen gissen naar de oorzaak van deze verschillen, die waarschijnlijk hun achtergrond
hebben in bestaande verschillen tussen instituties, cultuur en tradities hebben zij belangrijke
gevolgen voor het te voeren beleid. Zo suggereren de uitkomsten voor Nederland dat nieuwe
beleidsinitiatieven een grotere rol zouden moeten toekennen aan het verhogen van het
bewustzijn, terwijl in de VS meer nadruk op een goede informatievoorziening en een heldere
en eenvoudige presentatie van keuzeproblemen wellicht de meest effectieve aanpak is.
Terwijl de vier artikelen verschillende dimensies van financieel gedrag en individuele
beslissingen adresseren, is het overkoepelende beeld dat naar voren komt dat 1) financiële
vaardigheden van cruciaal belang zijn voor huishoudbeslissingen en 2) dat er een groot gat zit
tussen de daadwerkelijke financiële kennis van huishoudens aan de ene kant en de
vaardigheden die nodig zijn voor pensioenbeslissingen aan de andere kant. Nederlandse
werknemers lijken zich hiervan echter terdege bewust en zijn daarom niet happig op het
uitoefenen van meer invloed en verkrijgen van meer beleggingsverantwoordelijkheid ten
aanzien van het pensioen. De artikelen tonen verder dat het verminderen van de complexiteit
van financiële beslissingen de kwaliteit van huishoudbeslissingen ten goede kan komen; dat er
een belangrijke rol lijkt weggelegd voor informatievoorziening en educatie; en dat de manier
waarop keuzeproblemen zijn vormgegeven er toe doet voor financiële uitkomsten. Een
belangrijke vraag voor vervolgonderzoek is welke typen opleidingsprogramma’s en
activiteiten de financiële geletterdheid van gezinnen op de meest effectieve manier kunnen
verhogen.
Al met al, laten de resultaten echter zien dat het niet waarschijnlijk is dat financiële
educatie alleen voldoende zal zijn om moeilijkheden die gezinnen ondervinden bij het maken
van financiële beslissingen op te lossen. Het beleid moet er tevens op gericht zijn om waar
mogelijk financiële beslissingen eenvoudiger te maken en te waarborgen dat informatie en
advies over financiële producten begrijpelijk is en op een transparante, onafhankelijke manier
tot stand komt. Wat wel duidelijk is dat een hoger niveau van financiële kennis gepaard gaat
met veel voordelen in termen van wat algemeen als slim financieel gedrag wordt beschouwd.
188
Literatuur Ameriks, J., A. Caplin, en J. Leahy, 2003, Wealth accumulation and the propensity to plan, Quarterly Journal of Economics, 118, 1007-1047. Browning, M., en A. Lusardi, 1996, Household saving: Micro theories and micro facts, Journal of Economic Literature, 34, 1797-1855. Campbell, J., 2006, Household finance, Journal of Finance, 61, 1553-1604. Donkers, B, en A. van Soest, 1999, Subjective measures of household preferences and financial decisions, Journal of Economic Psychology, 20, 613-642. Friedman, M., 1957, A Theory of the Consumption Function, Princeton University Press. Guiso, L., M. Haliassos, en T. Jappelli, 2002, Household Portfolios, MIT Press, Cambridge. Haliassos, M., en C. Bertaut, 1995, Why do so few hold stocks?, Economic Journal, 105, 1110-1129. Hurd, M., en K. McGarry, 2002, The predictive validity of subjective probabilities of survival, Economic Journal, 112, 966-985. Kapteyn, A. en F. Teppa, 2002, Subjective measures of risk aversion and portfolio choice, mimeo, RAND. Lusardi, A., 2004, Saving and the effectiveness of financial education. In: O. Mitchell en S. Utkus (eds.), Pension Design and Structure: New Lessons from Behavioral Finance, Oxford University Press, 157-184. Lusardi, A., en O. Mitchell, 2007, Baby boomers retirement security: The role of planning, financial literacy and housing wealth, Journal of Monetary Economics, 54, 205-224. Manski, C., 2004, Measuring expectations, Econometrica, 72, 1329-1376. Modigliani, F., en R. Brumberg, 1954, Utility analysis and the consumption function: An interpretation of cross-section data. In: K. Kurihara (ed.), Post-Keynesian Economics, New Brunswick, New Jersey, Rutgers University Press, 388-436.
189
CURRICULUM VITAE
Maarten van Rooij (1970) works as a senior research economist in the
Economics & Research Division of De Nederlandsche Bank. He is
affiliated to Netspar (Network for Studies on Pensions, Aging and
Retirement) and member of the editorial board of the Dutch journal
TPEdigitaal. In 1995 he received his Masters degree in Econometrics
(cum laude) from Tilburg University with a special focus in the fields of
monetary economics, operations research and applied econometrics. He has worked among
others on macroeconomic forecasting and business cycle analysis at De Nederlandsche Bank
for about seven years before switching to the field of pensions and household financial
behavior in 2003.